linux/fs/ext4/fast_commit.c

2305 lines
64 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
/*
* fs/ext4/fast_commit.c
*
* Written by Harshad Shirwadkar <harshadshirwadkar@gmail.com>
*
* Ext4 fast commits routines.
*/
#include "ext4.h"
#include "ext4_jbd2.h"
#include "ext4_extents.h"
#include "mballoc.h"
/*
* Ext4 Fast Commits
* -----------------
*
* Ext4 fast commits implement fine grained journalling for Ext4.
*
* Fast commits are organized as a log of tag-length-value (TLV) structs. (See
* struct ext4_fc_tl). Each TLV contains some delta that is replayed TLV by
* TLV during the recovery phase. For the scenarios for which we currently
* don't have replay code, fast commit falls back to full commits.
* Fast commits record delta in one of the following three categories.
*
* (A) Directory entry updates:
*
* - EXT4_FC_TAG_UNLINK - records directory entry unlink
* - EXT4_FC_TAG_LINK - records directory entry link
* - EXT4_FC_TAG_CREAT - records inode and directory entry creation
*
* (B) File specific data range updates:
*
* - EXT4_FC_TAG_ADD_RANGE - records addition of new blocks to an inode
* - EXT4_FC_TAG_DEL_RANGE - records deletion of blocks from an inode
*
* (C) Inode metadata (mtime / ctime etc):
*
* - EXT4_FC_TAG_INODE - record the inode that should be replayed
* during recovery. Note that iblocks field is
* not replayed and instead derived during
* replay.
* Commit Operation
* ----------------
* With fast commits, we maintain all the directory entry operations in the
* order in which they are issued in an in-memory queue. This queue is flushed
* to disk during the commit operation. We also maintain a list of inodes
* that need to be committed during a fast commit in another in memory queue of
* inodes. During the commit operation, we commit in the following order:
*
* [1] Lock inodes for any further data updates by setting COMMITTING state
* [2] Submit data buffers of all the inodes
* [3] Wait for [2] to complete
* [4] Commit all the directory entry updates in the fast commit space
* [5] Commit all the changed inode structures
* [6] Write tail tag (this tag ensures the atomicity, please read the following
* section for more details).
* [7] Wait for [4], [5] and [6] to complete.
*
* All the inode updates must call ext4_fc_start_update() before starting an
* update. If such an ongoing update is present, fast commit waits for it to
* complete. The completion of such an update is marked by
* ext4_fc_stop_update().
*
* Fast Commit Ineligibility
* -------------------------
*
* Not all operations are supported by fast commits today (e.g extended
* attributes). Fast commit ineligibility is marked by calling
* ext4_fc_mark_ineligible(): This makes next fast commit operation to fall back
* to full commit.
*
* Atomicity of commits
* --------------------
* In order to guarantee atomicity during the commit operation, fast commit
* uses "EXT4_FC_TAG_TAIL" tag that marks a fast commit as complete. Tail
* tag contains CRC of the contents and TID of the transaction after which
* this fast commit should be applied. Recovery code replays fast commit
* logs only if there's at least 1 valid tail present. For every fast commit
* operation, there is 1 tail. This means, we may end up with multiple tails
* in the fast commit space. Here's an example:
*
* - Create a new file A and remove existing file B
* - fsync()
* - Append contents to file A
* - Truncate file A
* - fsync()
*
* The fast commit space at the end of above operations would look like this:
* [HEAD] [CREAT A] [UNLINK B] [TAIL] [ADD_RANGE A] [DEL_RANGE A] [TAIL]
* |<--- Fast Commit 1 --->|<--- Fast Commit 2 ---->|
*
* Replay code should thus check for all the valid tails in the FC area.
*
* Fast Commit Replay Idempotence
* ------------------------------
*
* Fast commits tags are idempotent in nature provided the recovery code follows
* certain rules. The guiding principle that the commit path follows while
* committing is that it stores the result of a particular operation instead of
* storing the procedure.
*
* Let's consider this rename operation: 'mv /a /b'. Let's assume dirent '/a'
* was associated with inode 10. During fast commit, instead of storing this
* operation as a procedure "rename a to b", we store the resulting file system
* state as a "series" of outcomes:
*
* - Link dirent b to inode 10
* - Unlink dirent a
* - Inode <10> with valid refcount
*
* Now when recovery code runs, it needs "enforce" this state on the file
* system. This is what guarantees idempotence of fast commit replay.
*
* Let's take an example of a procedure that is not idempotent and see how fast
* commits make it idempotent. Consider following sequence of operations:
*
* rm A; mv B A; read A
* (x) (y) (z)
*
* (x), (y) and (z) are the points at which we can crash. If we store this
* sequence of operations as is then the replay is not idempotent. Let's say
* while in replay, we crash at (z). During the second replay, file A (which was
* actually created as a result of "mv B A" operation) would get deleted. Thus,
* file named A would be absent when we try to read A. So, this sequence of
* operations is not idempotent. However, as mentioned above, instead of storing
* the procedure fast commits store the outcome of each procedure. Thus the fast
* commit log for above procedure would be as follows:
*
* (Let's assume dirent A was linked to inode 10 and dirent B was linked to
* inode 11 before the replay)
*
* [Unlink A] [Link A to inode 11] [Unlink B] [Inode 11]
* (w) (x) (y) (z)
*
* If we crash at (z), we will have file A linked to inode 11. During the second
* replay, we will remove file A (inode 11). But we will create it back and make
* it point to inode 11. We won't find B, so we'll just skip that step. At this
* point, the refcount for inode 11 is not reliable, but that gets fixed by the
* replay of last inode 11 tag. Crashes at points (w), (x) and (y) get handled
* similarly. Thus, by converting a non-idempotent procedure into a series of
* idempotent outcomes, fast commits ensured idempotence during the replay.
*
* TODOs
* -----
*
* 0) Fast commit replay path hardening: Fast commit replay code should use
* journal handles to make sure all the updates it does during the replay
* path are atomic. With that if we crash during fast commit replay, after
* trying to do recovery again, we will find a file system where fast commit
* area is invalid (because new full commit would be found). In order to deal
* with that, fast commit replay code should ensure that the "FC_REPLAY"
* superblock state is persisted before starting the replay, so that after
* the crash, fast commit recovery code can look at that flag and perform
* fast commit recovery even if that area is invalidated by later full
* commits.
*
* 1) Fast commit's commit path locks the entire file system during fast
* commit. This has significant performance penalty. Instead of that, we
* should use ext4_fc_start/stop_update functions to start inode level
* updates from ext4_journal_start/stop. Once we do that we can drop file
* system locking during commit path.
*
* 2) Handle more ineligible cases.
*/
#include <trace/events/ext4.h>
static struct kmem_cache *ext4_fc_dentry_cachep;
static void ext4_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
{
BUFFER_TRACE(bh, "");
if (uptodate) {
ext4_debug("%s: Block %lld up-to-date",
__func__, bh->b_blocknr);
set_buffer_uptodate(bh);
} else {
ext4_debug("%s: Block %lld not up-to-date",
__func__, bh->b_blocknr);
clear_buffer_uptodate(bh);
}
unlock_buffer(bh);
}
static inline void ext4_fc_reset_inode(struct inode *inode)
{
struct ext4_inode_info *ei = EXT4_I(inode);
ei->i_fc_lblk_start = 0;
ei->i_fc_lblk_len = 0;
}
void ext4_fc_init_inode(struct inode *inode)
{
struct ext4_inode_info *ei = EXT4_I(inode);
ext4_fc_reset_inode(inode);
ext4_clear_inode_state(inode, EXT4_STATE_FC_COMMITTING);
INIT_LIST_HEAD(&ei->i_fc_list);
ext4: improve fast_commit performance and scalability Currently ext4_fc_commit_dentry_updates() is of quadratic time complexity, which is causing performance bottlenecks with high threads/file/dir count with fs_mark. This patch makes commit dentry updates (and hence ext4_fc_commit()) path to linear time complexity. Hence improves the performance of workloads which does fsync on multiple threads/open files one-by-one. Absolute numbers in avg file creates per sec (from fs_mark in 1K order) ======================================================================= no. Order without-patch(K) with-patch(K) Diff(%) 1 1 16.90 17.51 +3.60 2 2,2 32.08 31.80 -0.87 3 3,3 53.97 55.01 +1.92 4 4,4 78.94 76.90 -2.58 5 5,5 95.82 95.37 -0.46 6 6,6 87.92 103.38 +17.58 7 6,10 0.73 126.13 +17178.08 8 6,14 2.33 143.19 +6045.49 workload type ============== For e.g. 7th row order of 6,10 (2^6 == 64 && 2^10 == 1024) echo /run/riteshh/mnt/{1..64} |sed -E 's/[[:space:]]+/ -d /g' \ | xargs -I {} bash -c "sudo fs_mark -L 100 -D 1024 -n 1024 -s0 -S5 -d {}" Perf profile (w/o patches) ============================= 87.15% [kernel] [k] ext4_fc_commit --> Heavy contention/bottleneck 1.98% [kernel] [k] perf_event_interrupt 0.96% [kernel] [k] power_pmu_enable 0.91% [kernel] [k] update_sd_lb_stats.constprop.0 0.67% [kernel] [k] ktime_get Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com> Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com> Link: https://lore.kernel.org/r/930f35d4fd5f83e2673c868781d9ebf15e91bf4e.1645426817.git.riteshh@linux.ibm.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-02-21 15:56:15 +08:00
INIT_LIST_HEAD(&ei->i_fc_dilist);
init_waitqueue_head(&ei->i_fc_wait);
atomic_set(&ei->i_fc_updates, 0);
}
/* This function must be called with sbi->s_fc_lock held. */
static void ext4_fc_wait_committing_inode(struct inode *inode)
__releases(&EXT4_SB(inode->i_sb)->s_fc_lock)
{
wait_queue_head_t *wq;
struct ext4_inode_info *ei = EXT4_I(inode);
#if (BITS_PER_LONG < 64)
DEFINE_WAIT_BIT(wait, &ei->i_state_flags,
EXT4_STATE_FC_COMMITTING);
wq = bit_waitqueue(&ei->i_state_flags,
EXT4_STATE_FC_COMMITTING);
#else
DEFINE_WAIT_BIT(wait, &ei->i_flags,
EXT4_STATE_FC_COMMITTING);
wq = bit_waitqueue(&ei->i_flags,
EXT4_STATE_FC_COMMITTING);
#endif
lockdep_assert_held(&EXT4_SB(inode->i_sb)->s_fc_lock);
prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
schedule();
finish_wait(wq, &wait.wq_entry);
}
static bool ext4_fc_disabled(struct super_block *sb)
{
return (!test_opt2(sb, JOURNAL_FAST_COMMIT) ||
(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY));
}
/*
* Inform Ext4's fast about start of an inode update
*
* This function is called by the high level call VFS callbacks before
* performing any inode update. This function blocks if there's an ongoing
* fast commit on the inode in question.
*/
void ext4_fc_start_update(struct inode *inode)
{
struct ext4_inode_info *ei = EXT4_I(inode);
if (ext4_fc_disabled(inode->i_sb))
return;
restart:
spin_lock(&EXT4_SB(inode->i_sb)->s_fc_lock);
if (list_empty(&ei->i_fc_list))
goto out;
if (ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING)) {
ext4_fc_wait_committing_inode(inode);
goto restart;
}
out:
atomic_inc(&ei->i_fc_updates);
spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
}
/*
* Stop inode update and wake up waiting fast commits if any.
*/
void ext4_fc_stop_update(struct inode *inode)
{
struct ext4_inode_info *ei = EXT4_I(inode);
if (ext4_fc_disabled(inode->i_sb))
return;
if (atomic_dec_and_test(&ei->i_fc_updates))
wake_up_all(&ei->i_fc_wait);
}
/*
* Remove inode from fast commit list. If the inode is being committed
* we wait until inode commit is done.
*/
void ext4_fc_del(struct inode *inode)
{
struct ext4_inode_info *ei = EXT4_I(inode);
ext4: improve fast_commit performance and scalability Currently ext4_fc_commit_dentry_updates() is of quadratic time complexity, which is causing performance bottlenecks with high threads/file/dir count with fs_mark. This patch makes commit dentry updates (and hence ext4_fc_commit()) path to linear time complexity. Hence improves the performance of workloads which does fsync on multiple threads/open files one-by-one. Absolute numbers in avg file creates per sec (from fs_mark in 1K order) ======================================================================= no. Order without-patch(K) with-patch(K) Diff(%) 1 1 16.90 17.51 +3.60 2 2,2 32.08 31.80 -0.87 3 3,3 53.97 55.01 +1.92 4 4,4 78.94 76.90 -2.58 5 5,5 95.82 95.37 -0.46 6 6,6 87.92 103.38 +17.58 7 6,10 0.73 126.13 +17178.08 8 6,14 2.33 143.19 +6045.49 workload type ============== For e.g. 7th row order of 6,10 (2^6 == 64 && 2^10 == 1024) echo /run/riteshh/mnt/{1..64} |sed -E 's/[[:space:]]+/ -d /g' \ | xargs -I {} bash -c "sudo fs_mark -L 100 -D 1024 -n 1024 -s0 -S5 -d {}" Perf profile (w/o patches) ============================= 87.15% [kernel] [k] ext4_fc_commit --> Heavy contention/bottleneck 1.98% [kernel] [k] perf_event_interrupt 0.96% [kernel] [k] power_pmu_enable 0.91% [kernel] [k] update_sd_lb_stats.constprop.0 0.67% [kernel] [k] ktime_get Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com> Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com> Link: https://lore.kernel.org/r/930f35d4fd5f83e2673c868781d9ebf15e91bf4e.1645426817.git.riteshh@linux.ibm.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-02-21 15:56:15 +08:00
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
struct ext4_fc_dentry_update *fc_dentry;
if (ext4_fc_disabled(inode->i_sb))
return;
restart:
spin_lock(&EXT4_SB(inode->i_sb)->s_fc_lock);
ext4: improve fast_commit performance and scalability Currently ext4_fc_commit_dentry_updates() is of quadratic time complexity, which is causing performance bottlenecks with high threads/file/dir count with fs_mark. This patch makes commit dentry updates (and hence ext4_fc_commit()) path to linear time complexity. Hence improves the performance of workloads which does fsync on multiple threads/open files one-by-one. Absolute numbers in avg file creates per sec (from fs_mark in 1K order) ======================================================================= no. Order without-patch(K) with-patch(K) Diff(%) 1 1 16.90 17.51 +3.60 2 2,2 32.08 31.80 -0.87 3 3,3 53.97 55.01 +1.92 4 4,4 78.94 76.90 -2.58 5 5,5 95.82 95.37 -0.46 6 6,6 87.92 103.38 +17.58 7 6,10 0.73 126.13 +17178.08 8 6,14 2.33 143.19 +6045.49 workload type ============== For e.g. 7th row order of 6,10 (2^6 == 64 && 2^10 == 1024) echo /run/riteshh/mnt/{1..64} |sed -E 's/[[:space:]]+/ -d /g' \ | xargs -I {} bash -c "sudo fs_mark -L 100 -D 1024 -n 1024 -s0 -S5 -d {}" Perf profile (w/o patches) ============================= 87.15% [kernel] [k] ext4_fc_commit --> Heavy contention/bottleneck 1.98% [kernel] [k] perf_event_interrupt 0.96% [kernel] [k] power_pmu_enable 0.91% [kernel] [k] update_sd_lb_stats.constprop.0 0.67% [kernel] [k] ktime_get Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com> Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com> Link: https://lore.kernel.org/r/930f35d4fd5f83e2673c868781d9ebf15e91bf4e.1645426817.git.riteshh@linux.ibm.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-02-21 15:56:15 +08:00
if (list_empty(&ei->i_fc_list) && list_empty(&ei->i_fc_dilist)) {
spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
return;
}
if (ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING)) {
ext4_fc_wait_committing_inode(inode);
goto restart;
}
ext4: improve fast_commit performance and scalability Currently ext4_fc_commit_dentry_updates() is of quadratic time complexity, which is causing performance bottlenecks with high threads/file/dir count with fs_mark. This patch makes commit dentry updates (and hence ext4_fc_commit()) path to linear time complexity. Hence improves the performance of workloads which does fsync on multiple threads/open files one-by-one. Absolute numbers in avg file creates per sec (from fs_mark in 1K order) ======================================================================= no. Order without-patch(K) with-patch(K) Diff(%) 1 1 16.90 17.51 +3.60 2 2,2 32.08 31.80 -0.87 3 3,3 53.97 55.01 +1.92 4 4,4 78.94 76.90 -2.58 5 5,5 95.82 95.37 -0.46 6 6,6 87.92 103.38 +17.58 7 6,10 0.73 126.13 +17178.08 8 6,14 2.33 143.19 +6045.49 workload type ============== For e.g. 7th row order of 6,10 (2^6 == 64 && 2^10 == 1024) echo /run/riteshh/mnt/{1..64} |sed -E 's/[[:space:]]+/ -d /g' \ | xargs -I {} bash -c "sudo fs_mark -L 100 -D 1024 -n 1024 -s0 -S5 -d {}" Perf profile (w/o patches) ============================= 87.15% [kernel] [k] ext4_fc_commit --> Heavy contention/bottleneck 1.98% [kernel] [k] perf_event_interrupt 0.96% [kernel] [k] power_pmu_enable 0.91% [kernel] [k] update_sd_lb_stats.constprop.0 0.67% [kernel] [k] ktime_get Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com> Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com> Link: https://lore.kernel.org/r/930f35d4fd5f83e2673c868781d9ebf15e91bf4e.1645426817.git.riteshh@linux.ibm.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-02-21 15:56:15 +08:00
if (!list_empty(&ei->i_fc_list))
list_del_init(&ei->i_fc_list);
/*
* Since this inode is getting removed, let's also remove all FC
* dentry create references, since it is not needed to log it anyways.
*/
if (list_empty(&ei->i_fc_dilist)) {
spin_unlock(&sbi->s_fc_lock);
return;
}
fc_dentry = list_first_entry(&ei->i_fc_dilist, struct ext4_fc_dentry_update, fcd_dilist);
WARN_ON(fc_dentry->fcd_op != EXT4_FC_TAG_CREAT);
list_del_init(&fc_dentry->fcd_list);
list_del_init(&fc_dentry->fcd_dilist);
WARN_ON(!list_empty(&ei->i_fc_dilist));
spin_unlock(&sbi->s_fc_lock);
if (fc_dentry->fcd_name.name &&
fc_dentry->fcd_name.len > DNAME_INLINE_LEN)
kfree(fc_dentry->fcd_name.name);
kmem_cache_free(ext4_fc_dentry_cachep, fc_dentry);
return;
}
/*
* Mark file system as fast commit ineligible, and record latest
* ineligible transaction tid. This means until the recorded
* transaction, commit operation would result in a full jbd2 commit.
*/
void ext4_fc_mark_ineligible(struct super_block *sb, int reason, handle_t *handle)
{
struct ext4_sb_info *sbi = EXT4_SB(sb);
tid_t tid;
if (ext4_fc_disabled(sb))
return;
ext4_set_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
if (handle && !IS_ERR(handle))
tid = handle->h_transaction->t_tid;
else {
read_lock(&sbi->s_journal->j_state_lock);
tid = sbi->s_journal->j_running_transaction ?
sbi->s_journal->j_running_transaction->t_tid : 0;
read_unlock(&sbi->s_journal->j_state_lock);
}
spin_lock(&sbi->s_fc_lock);
if (sbi->s_fc_ineligible_tid < tid)
sbi->s_fc_ineligible_tid = tid;
spin_unlock(&sbi->s_fc_lock);
WARN_ON(reason >= EXT4_FC_REASON_MAX);
sbi->s_fc_stats.fc_ineligible_reason_count[reason]++;
}
/*
* Generic fast commit tracking function. If this is the first time this we are
* called after a full commit, we initialize fast commit fields and then call
* __fc_track_fn() with update = 0. If we have already been called after a full
* commit, we pass update = 1. Based on that, the track function can determine
* if it needs to track a field for the first time or if it needs to just
* update the previously tracked value.
*
* If enqueue is set, this function enqueues the inode in fast commit list.
*/
static int ext4_fc_track_template(
handle_t *handle, struct inode *inode,
int (*__fc_track_fn)(struct inode *, void *, bool),
void *args, int enqueue)
{
bool update = false;
struct ext4_inode_info *ei = EXT4_I(inode);
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
tid_t tid = 0;
int ret;
tid = handle->h_transaction->t_tid;
mutex_lock(&ei->i_fc_lock);
if (tid == ei->i_sync_tid) {
update = true;
} else {
ext4_fc_reset_inode(inode);
ei->i_sync_tid = tid;
}
ret = __fc_track_fn(inode, args, update);
mutex_unlock(&ei->i_fc_lock);
if (!enqueue)
return ret;
spin_lock(&sbi->s_fc_lock);
if (list_empty(&EXT4_I(inode)->i_fc_list))
list_add_tail(&EXT4_I(inode)->i_fc_list,
(sbi->s_journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
sbi->s_journal->j_flags & JBD2_FAST_COMMIT_ONGOING) ?
&sbi->s_fc_q[FC_Q_STAGING] :
&sbi->s_fc_q[FC_Q_MAIN]);
spin_unlock(&sbi->s_fc_lock);
return ret;
}
struct __track_dentry_update_args {
struct dentry *dentry;
int op;
};
/* __track_fn for directory entry updates. Called with ei->i_fc_lock. */
static int __track_dentry_update(struct inode *inode, void *arg, bool update)
{
struct ext4_fc_dentry_update *node;
struct ext4_inode_info *ei = EXT4_I(inode);
struct __track_dentry_update_args *dentry_update =
(struct __track_dentry_update_args *)arg;
struct dentry *dentry = dentry_update->dentry;
struct inode *dir = dentry->d_parent->d_inode;
struct super_block *sb = inode->i_sb;
struct ext4_sb_info *sbi = EXT4_SB(sb);
mutex_unlock(&ei->i_fc_lock);
if (IS_ENCRYPTED(dir)) {
ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_ENCRYPTED_FILENAME,
NULL);
mutex_lock(&ei->i_fc_lock);
return -EOPNOTSUPP;
}
node = kmem_cache_alloc(ext4_fc_dentry_cachep, GFP_NOFS);
if (!node) {
ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_NOMEM, NULL);
mutex_lock(&ei->i_fc_lock);
return -ENOMEM;
}
node->fcd_op = dentry_update->op;
node->fcd_parent = dir->i_ino;
node->fcd_ino = inode->i_ino;
if (dentry->d_name.len > DNAME_INLINE_LEN) {
node->fcd_name.name = kmalloc(dentry->d_name.len, GFP_NOFS);
if (!node->fcd_name.name) {
kmem_cache_free(ext4_fc_dentry_cachep, node);
ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_NOMEM, NULL);
mutex_lock(&ei->i_fc_lock);
return -ENOMEM;
}
memcpy((u8 *)node->fcd_name.name, dentry->d_name.name,
dentry->d_name.len);
} else {
memcpy(node->fcd_iname, dentry->d_name.name,
dentry->d_name.len);
node->fcd_name.name = node->fcd_iname;
}
node->fcd_name.len = dentry->d_name.len;
ext4: improve fast_commit performance and scalability Currently ext4_fc_commit_dentry_updates() is of quadratic time complexity, which is causing performance bottlenecks with high threads/file/dir count with fs_mark. This patch makes commit dentry updates (and hence ext4_fc_commit()) path to linear time complexity. Hence improves the performance of workloads which does fsync on multiple threads/open files one-by-one. Absolute numbers in avg file creates per sec (from fs_mark in 1K order) ======================================================================= no. Order without-patch(K) with-patch(K) Diff(%) 1 1 16.90 17.51 +3.60 2 2,2 32.08 31.80 -0.87 3 3,3 53.97 55.01 +1.92 4 4,4 78.94 76.90 -2.58 5 5,5 95.82 95.37 -0.46 6 6,6 87.92 103.38 +17.58 7 6,10 0.73 126.13 +17178.08 8 6,14 2.33 143.19 +6045.49 workload type ============== For e.g. 7th row order of 6,10 (2^6 == 64 && 2^10 == 1024) echo /run/riteshh/mnt/{1..64} |sed -E 's/[[:space:]]+/ -d /g' \ | xargs -I {} bash -c "sudo fs_mark -L 100 -D 1024 -n 1024 -s0 -S5 -d {}" Perf profile (w/o patches) ============================= 87.15% [kernel] [k] ext4_fc_commit --> Heavy contention/bottleneck 1.98% [kernel] [k] perf_event_interrupt 0.96% [kernel] [k] power_pmu_enable 0.91% [kernel] [k] update_sd_lb_stats.constprop.0 0.67% [kernel] [k] ktime_get Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com> Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com> Link: https://lore.kernel.org/r/930f35d4fd5f83e2673c868781d9ebf15e91bf4e.1645426817.git.riteshh@linux.ibm.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-02-21 15:56:15 +08:00
INIT_LIST_HEAD(&node->fcd_dilist);
spin_lock(&sbi->s_fc_lock);
if (sbi->s_journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
sbi->s_journal->j_flags & JBD2_FAST_COMMIT_ONGOING)
list_add_tail(&node->fcd_list,
&sbi->s_fc_dentry_q[FC_Q_STAGING]);
else
list_add_tail(&node->fcd_list, &sbi->s_fc_dentry_q[FC_Q_MAIN]);
ext4: improve fast_commit performance and scalability Currently ext4_fc_commit_dentry_updates() is of quadratic time complexity, which is causing performance bottlenecks with high threads/file/dir count with fs_mark. This patch makes commit dentry updates (and hence ext4_fc_commit()) path to linear time complexity. Hence improves the performance of workloads which does fsync on multiple threads/open files one-by-one. Absolute numbers in avg file creates per sec (from fs_mark in 1K order) ======================================================================= no. Order without-patch(K) with-patch(K) Diff(%) 1 1 16.90 17.51 +3.60 2 2,2 32.08 31.80 -0.87 3 3,3 53.97 55.01 +1.92 4 4,4 78.94 76.90 -2.58 5 5,5 95.82 95.37 -0.46 6 6,6 87.92 103.38 +17.58 7 6,10 0.73 126.13 +17178.08 8 6,14 2.33 143.19 +6045.49 workload type ============== For e.g. 7th row order of 6,10 (2^6 == 64 && 2^10 == 1024) echo /run/riteshh/mnt/{1..64} |sed -E 's/[[:space:]]+/ -d /g' \ | xargs -I {} bash -c "sudo fs_mark -L 100 -D 1024 -n 1024 -s0 -S5 -d {}" Perf profile (w/o patches) ============================= 87.15% [kernel] [k] ext4_fc_commit --> Heavy contention/bottleneck 1.98% [kernel] [k] perf_event_interrupt 0.96% [kernel] [k] power_pmu_enable 0.91% [kernel] [k] update_sd_lb_stats.constprop.0 0.67% [kernel] [k] ktime_get Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com> Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com> Link: https://lore.kernel.org/r/930f35d4fd5f83e2673c868781d9ebf15e91bf4e.1645426817.git.riteshh@linux.ibm.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-02-21 15:56:15 +08:00
/*
* This helps us keep a track of all fc_dentry updates which is part of
* this ext4 inode. So in case the inode is getting unlinked, before
* even we get a chance to fsync, we could remove all fc_dentry
* references while evicting the inode in ext4_fc_del().
* Also with this, we don't need to loop over all the inodes in
* sbi->s_fc_q to get the corresponding inode in
* ext4_fc_commit_dentry_updates().
*/
if (dentry_update->op == EXT4_FC_TAG_CREAT) {
WARN_ON(!list_empty(&ei->i_fc_dilist));
list_add_tail(&node->fcd_dilist, &ei->i_fc_dilist);
}
spin_unlock(&sbi->s_fc_lock);
mutex_lock(&ei->i_fc_lock);
return 0;
}
void __ext4_fc_track_unlink(handle_t *handle,
struct inode *inode, struct dentry *dentry)
{
struct __track_dentry_update_args args;
int ret;
args.dentry = dentry;
args.op = EXT4_FC_TAG_UNLINK;
ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
(void *)&args, 0);
trace_ext4_fc_track_unlink(handle, inode, dentry, ret);
}
void ext4_fc_track_unlink(handle_t *handle, struct dentry *dentry)
{
struct inode *inode = d_inode(dentry);
if (ext4_fc_disabled(inode->i_sb))
return;
if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
return;
__ext4_fc_track_unlink(handle, inode, dentry);
}
void __ext4_fc_track_link(handle_t *handle,
struct inode *inode, struct dentry *dentry)
{
struct __track_dentry_update_args args;
int ret;
args.dentry = dentry;
args.op = EXT4_FC_TAG_LINK;
ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
(void *)&args, 0);
trace_ext4_fc_track_link(handle, inode, dentry, ret);
}
void ext4_fc_track_link(handle_t *handle, struct dentry *dentry)
{
struct inode *inode = d_inode(dentry);
if (ext4_fc_disabled(inode->i_sb))
return;
if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
return;
__ext4_fc_track_link(handle, inode, dentry);
}
ext4: fix rename whiteout with fast commit This patch adds rename whiteout support in fast commits. Note that the whiteout object that gets created is actually char device. Which imples, the function ext4_inode_journal_mode(struct inode *inode) would return "JOURNAL_DATA" for this inode. This has a consequence in fast commit code that it will make creation of the whiteout object a fast-commit ineligible behavior and thus will fall back to full commits. With this patch, this can be observed by running fast commits with rename whiteout and seeing the stats generated by ext4_fc_stats tracepoint as follows: ext4_fc_stats: dev 254:32 fc ineligible reasons: XATTR:0, CROSS_RENAME:0, JOURNAL_FLAG_CHANGE:0, NO_MEM:0, SWAP_BOOT:0, RESIZE:0, RENAME_DIR:0, FALLOC_RANGE:0, INODE_JOURNAL_DATA:16; num_commits:6, ineligible: 6, numblks: 3 So in short, this patch guarantees that in case of rename whiteout, we fall back to full commits. Amir mentioned that instead of creating a new whiteout object for every rename, we can create a static whiteout object with irrelevant nlink. That will make fast commits to not fall back to full commit. But until this happens, this patch will ensure correctness by falling back to full commits. Fixes: 8016e29f4362 ("ext4: fast commit recovery path") Cc: stable@kernel.org Signed-off-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com> Link: https://lore.kernel.org/r/20210316221921.1124955-1-harshadshirwadkar@gmail.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-03-17 06:19:21 +08:00
void __ext4_fc_track_create(handle_t *handle, struct inode *inode,
struct dentry *dentry)
{
struct __track_dentry_update_args args;
int ret;
args.dentry = dentry;
args.op = EXT4_FC_TAG_CREAT;
ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
(void *)&args, 0);
trace_ext4_fc_track_create(handle, inode, dentry, ret);
}
ext4: fix rename whiteout with fast commit This patch adds rename whiteout support in fast commits. Note that the whiteout object that gets created is actually char device. Which imples, the function ext4_inode_journal_mode(struct inode *inode) would return "JOURNAL_DATA" for this inode. This has a consequence in fast commit code that it will make creation of the whiteout object a fast-commit ineligible behavior and thus will fall back to full commits. With this patch, this can be observed by running fast commits with rename whiteout and seeing the stats generated by ext4_fc_stats tracepoint as follows: ext4_fc_stats: dev 254:32 fc ineligible reasons: XATTR:0, CROSS_RENAME:0, JOURNAL_FLAG_CHANGE:0, NO_MEM:0, SWAP_BOOT:0, RESIZE:0, RENAME_DIR:0, FALLOC_RANGE:0, INODE_JOURNAL_DATA:16; num_commits:6, ineligible: 6, numblks: 3 So in short, this patch guarantees that in case of rename whiteout, we fall back to full commits. Amir mentioned that instead of creating a new whiteout object for every rename, we can create a static whiteout object with irrelevant nlink. That will make fast commits to not fall back to full commit. But until this happens, this patch will ensure correctness by falling back to full commits. Fixes: 8016e29f4362 ("ext4: fast commit recovery path") Cc: stable@kernel.org Signed-off-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com> Link: https://lore.kernel.org/r/20210316221921.1124955-1-harshadshirwadkar@gmail.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-03-17 06:19:21 +08:00
void ext4_fc_track_create(handle_t *handle, struct dentry *dentry)
{
struct inode *inode = d_inode(dentry);
if (ext4_fc_disabled(inode->i_sb))
return;
if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
return;
__ext4_fc_track_create(handle, inode, dentry);
ext4: fix rename whiteout with fast commit This patch adds rename whiteout support in fast commits. Note that the whiteout object that gets created is actually char device. Which imples, the function ext4_inode_journal_mode(struct inode *inode) would return "JOURNAL_DATA" for this inode. This has a consequence in fast commit code that it will make creation of the whiteout object a fast-commit ineligible behavior and thus will fall back to full commits. With this patch, this can be observed by running fast commits with rename whiteout and seeing the stats generated by ext4_fc_stats tracepoint as follows: ext4_fc_stats: dev 254:32 fc ineligible reasons: XATTR:0, CROSS_RENAME:0, JOURNAL_FLAG_CHANGE:0, NO_MEM:0, SWAP_BOOT:0, RESIZE:0, RENAME_DIR:0, FALLOC_RANGE:0, INODE_JOURNAL_DATA:16; num_commits:6, ineligible: 6, numblks: 3 So in short, this patch guarantees that in case of rename whiteout, we fall back to full commits. Amir mentioned that instead of creating a new whiteout object for every rename, we can create a static whiteout object with irrelevant nlink. That will make fast commits to not fall back to full commit. But until this happens, this patch will ensure correctness by falling back to full commits. Fixes: 8016e29f4362 ("ext4: fast commit recovery path") Cc: stable@kernel.org Signed-off-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com> Link: https://lore.kernel.org/r/20210316221921.1124955-1-harshadshirwadkar@gmail.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-03-17 06:19:21 +08:00
}
/* __track_fn for inode tracking */
static int __track_inode(struct inode *inode, void *arg, bool update)
{
if (update)
return -EEXIST;
EXT4_I(inode)->i_fc_lblk_len = 0;
return 0;
}
void ext4_fc_track_inode(handle_t *handle, struct inode *inode)
{
int ret;
if (S_ISDIR(inode->i_mode))
return;
if (ext4_fc_disabled(inode->i_sb))
return;
if (ext4_should_journal_data(inode)) {
ext4_fc_mark_ineligible(inode->i_sb,
EXT4_FC_REASON_INODE_JOURNAL_DATA, handle);
return;
}
if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
return;
ret = ext4_fc_track_template(handle, inode, __track_inode, NULL, 1);
trace_ext4_fc_track_inode(handle, inode, ret);
}
struct __track_range_args {
ext4_lblk_t start, end;
};
/* __track_fn for tracking data updates */
static int __track_range(struct inode *inode, void *arg, bool update)
{
struct ext4_inode_info *ei = EXT4_I(inode);
ext4_lblk_t oldstart;
struct __track_range_args *__arg =
(struct __track_range_args *)arg;
if (inode->i_ino < EXT4_FIRST_INO(inode->i_sb)) {
ext4_debug("Special inode %ld being modified\n", inode->i_ino);
return -ECANCELED;
}
oldstart = ei->i_fc_lblk_start;
if (update && ei->i_fc_lblk_len > 0) {
ei->i_fc_lblk_start = min(ei->i_fc_lblk_start, __arg->start);
ei->i_fc_lblk_len =
max(oldstart + ei->i_fc_lblk_len - 1, __arg->end) -
ei->i_fc_lblk_start + 1;
} else {
ei->i_fc_lblk_start = __arg->start;
ei->i_fc_lblk_len = __arg->end - __arg->start + 1;
}
return 0;
}
void ext4_fc_track_range(handle_t *handle, struct inode *inode, ext4_lblk_t start,
ext4_lblk_t end)
{
struct __track_range_args args;
int ret;
if (S_ISDIR(inode->i_mode))
return;
if (ext4_fc_disabled(inode->i_sb))
return;
if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
return;
args.start = start;
args.end = end;
ret = ext4_fc_track_template(handle, inode, __track_range, &args, 1);
trace_ext4_fc_track_range(handle, inode, start, end, ret);
}
static void ext4_fc_submit_bh(struct super_block *sb, bool is_tail)
{
blk_opf_t write_flags = REQ_SYNC;
struct buffer_head *bh = EXT4_SB(sb)->s_fc_bh;
/* Add REQ_FUA | REQ_PREFLUSH only its tail */
if (test_opt(sb, BARRIER) && is_tail)
write_flags |= REQ_FUA | REQ_PREFLUSH;
lock_buffer(bh);
set_buffer_dirty(bh);
set_buffer_uptodate(bh);
bh->b_end_io = ext4_end_buffer_io_sync;
submit_bh(REQ_OP_WRITE | write_flags, bh);
EXT4_SB(sb)->s_fc_bh = NULL;
}
/* Ext4 commit path routines */
/*
* Allocate len bytes on a fast commit buffer.
*
* During the commit time this function is used to manage fast commit
* block space. We don't split a fast commit log onto different
* blocks. So this function makes sure that if there's not enough space
* on the current block, the remaining space in the current block is
* marked as unused by adding EXT4_FC_TAG_PAD tag. In that case,
* new block is from jbd2 and CRC is updated to reflect the padding
* we added.
*/
static u8 *ext4_fc_reserve_space(struct super_block *sb, int len, u32 *crc)
{
struct ext4_fc_tl tl;
struct ext4_sb_info *sbi = EXT4_SB(sb);
struct buffer_head *bh;
int bsize = sbi->s_journal->j_blocksize;
int ret, off = sbi->s_fc_bytes % bsize;
ext4: fix off-by-one errors in fast-commit block filling Due to several different off-by-one errors, or perhaps due to a late change in design that wasn't fully reflected in the code that was actually merged, there are several very strange constraints on how fast-commit blocks are filled with tlv entries: - tlvs must start at least 10 bytes before the end of the block, even though the minimum tlv length is 8. Otherwise, the replay code will ignore them. (BUG: ext4_fc_reserve_space() could violate this requirement if called with a len of blocksize - 9 or blocksize - 8. Fortunately, this doesn't seem to happen currently.) - tlvs must end at least 1 byte before the end of the block. Otherwise the replay code will consider them to be invalid. This quirk contributed to a bug (fixed by an earlier commit) where uninitialized memory was being leaked to disk in the last byte of blocks. Also, strangely these constraints don't apply to the replay code in e2fsprogs, which will accept any tlvs in the blocks (with no bounds checks at all, but that is a separate issue...). Given that this all seems to be a bug, let's fix it by just filling blocks with tlv entries in the natural way. Note that old kernels will be unable to replay fast-commit journals created by kernels that have this commit. Fixes: aa75f4d3daae ("ext4: main fast-commit commit path") Cc: <stable@vger.kernel.org> # v5.10+ Signed-off-by: Eric Biggers <ebiggers@google.com> Link: https://lore.kernel.org/r/20221106224841.279231-7-ebiggers@kernel.org Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-11-07 06:48:40 +08:00
int remaining;
u8 *dst;
/*
ext4: fix off-by-one errors in fast-commit block filling Due to several different off-by-one errors, or perhaps due to a late change in design that wasn't fully reflected in the code that was actually merged, there are several very strange constraints on how fast-commit blocks are filled with tlv entries: - tlvs must start at least 10 bytes before the end of the block, even though the minimum tlv length is 8. Otherwise, the replay code will ignore them. (BUG: ext4_fc_reserve_space() could violate this requirement if called with a len of blocksize - 9 or blocksize - 8. Fortunately, this doesn't seem to happen currently.) - tlvs must end at least 1 byte before the end of the block. Otherwise the replay code will consider them to be invalid. This quirk contributed to a bug (fixed by an earlier commit) where uninitialized memory was being leaked to disk in the last byte of blocks. Also, strangely these constraints don't apply to the replay code in e2fsprogs, which will accept any tlvs in the blocks (with no bounds checks at all, but that is a separate issue...). Given that this all seems to be a bug, let's fix it by just filling blocks with tlv entries in the natural way. Note that old kernels will be unable to replay fast-commit journals created by kernels that have this commit. Fixes: aa75f4d3daae ("ext4: main fast-commit commit path") Cc: <stable@vger.kernel.org> # v5.10+ Signed-off-by: Eric Biggers <ebiggers@google.com> Link: https://lore.kernel.org/r/20221106224841.279231-7-ebiggers@kernel.org Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-11-07 06:48:40 +08:00
* If 'len' is too long to fit in any block alongside a PAD tlv, then we
* cannot fulfill the request.
*/
ext4: fix off-by-one errors in fast-commit block filling Due to several different off-by-one errors, or perhaps due to a late change in design that wasn't fully reflected in the code that was actually merged, there are several very strange constraints on how fast-commit blocks are filled with tlv entries: - tlvs must start at least 10 bytes before the end of the block, even though the minimum tlv length is 8. Otherwise, the replay code will ignore them. (BUG: ext4_fc_reserve_space() could violate this requirement if called with a len of blocksize - 9 or blocksize - 8. Fortunately, this doesn't seem to happen currently.) - tlvs must end at least 1 byte before the end of the block. Otherwise the replay code will consider them to be invalid. This quirk contributed to a bug (fixed by an earlier commit) where uninitialized memory was being leaked to disk in the last byte of blocks. Also, strangely these constraints don't apply to the replay code in e2fsprogs, which will accept any tlvs in the blocks (with no bounds checks at all, but that is a separate issue...). Given that this all seems to be a bug, let's fix it by just filling blocks with tlv entries in the natural way. Note that old kernels will be unable to replay fast-commit journals created by kernels that have this commit. Fixes: aa75f4d3daae ("ext4: main fast-commit commit path") Cc: <stable@vger.kernel.org> # v5.10+ Signed-off-by: Eric Biggers <ebiggers@google.com> Link: https://lore.kernel.org/r/20221106224841.279231-7-ebiggers@kernel.org Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-11-07 06:48:40 +08:00
if (len > bsize - EXT4_FC_TAG_BASE_LEN)
return NULL;
ext4: fix off-by-one errors in fast-commit block filling Due to several different off-by-one errors, or perhaps due to a late change in design that wasn't fully reflected in the code that was actually merged, there are several very strange constraints on how fast-commit blocks are filled with tlv entries: - tlvs must start at least 10 bytes before the end of the block, even though the minimum tlv length is 8. Otherwise, the replay code will ignore them. (BUG: ext4_fc_reserve_space() could violate this requirement if called with a len of blocksize - 9 or blocksize - 8. Fortunately, this doesn't seem to happen currently.) - tlvs must end at least 1 byte before the end of the block. Otherwise the replay code will consider them to be invalid. This quirk contributed to a bug (fixed by an earlier commit) where uninitialized memory was being leaked to disk in the last byte of blocks. Also, strangely these constraints don't apply to the replay code in e2fsprogs, which will accept any tlvs in the blocks (with no bounds checks at all, but that is a separate issue...). Given that this all seems to be a bug, let's fix it by just filling blocks with tlv entries in the natural way. Note that old kernels will be unable to replay fast-commit journals created by kernels that have this commit. Fixes: aa75f4d3daae ("ext4: main fast-commit commit path") Cc: <stable@vger.kernel.org> # v5.10+ Signed-off-by: Eric Biggers <ebiggers@google.com> Link: https://lore.kernel.org/r/20221106224841.279231-7-ebiggers@kernel.org Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-11-07 06:48:40 +08:00
if (!sbi->s_fc_bh) {
ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
if (ret)
return NULL;
sbi->s_fc_bh = bh;
}
dst = sbi->s_fc_bh->b_data + off;
ext4: fix off-by-one errors in fast-commit block filling Due to several different off-by-one errors, or perhaps due to a late change in design that wasn't fully reflected in the code that was actually merged, there are several very strange constraints on how fast-commit blocks are filled with tlv entries: - tlvs must start at least 10 bytes before the end of the block, even though the minimum tlv length is 8. Otherwise, the replay code will ignore them. (BUG: ext4_fc_reserve_space() could violate this requirement if called with a len of blocksize - 9 or blocksize - 8. Fortunately, this doesn't seem to happen currently.) - tlvs must end at least 1 byte before the end of the block. Otherwise the replay code will consider them to be invalid. This quirk contributed to a bug (fixed by an earlier commit) where uninitialized memory was being leaked to disk in the last byte of blocks. Also, strangely these constraints don't apply to the replay code in e2fsprogs, which will accept any tlvs in the blocks (with no bounds checks at all, but that is a separate issue...). Given that this all seems to be a bug, let's fix it by just filling blocks with tlv entries in the natural way. Note that old kernels will be unable to replay fast-commit journals created by kernels that have this commit. Fixes: aa75f4d3daae ("ext4: main fast-commit commit path") Cc: <stable@vger.kernel.org> # v5.10+ Signed-off-by: Eric Biggers <ebiggers@google.com> Link: https://lore.kernel.org/r/20221106224841.279231-7-ebiggers@kernel.org Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-11-07 06:48:40 +08:00
/*
* Allocate the bytes in the current block if we can do so while still
* leaving enough space for a PAD tlv.
*/
remaining = bsize - EXT4_FC_TAG_BASE_LEN - off;
if (len <= remaining) {
sbi->s_fc_bytes += len;
return dst;
}
/*
* Else, terminate the current block with a PAD tlv, then allocate a new
* block and allocate the bytes at the start of that new block.
*/
tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_PAD);
ext4: fix off-by-one errors in fast-commit block filling Due to several different off-by-one errors, or perhaps due to a late change in design that wasn't fully reflected in the code that was actually merged, there are several very strange constraints on how fast-commit blocks are filled with tlv entries: - tlvs must start at least 10 bytes before the end of the block, even though the minimum tlv length is 8. Otherwise, the replay code will ignore them. (BUG: ext4_fc_reserve_space() could violate this requirement if called with a len of blocksize - 9 or blocksize - 8. Fortunately, this doesn't seem to happen currently.) - tlvs must end at least 1 byte before the end of the block. Otherwise the replay code will consider them to be invalid. This quirk contributed to a bug (fixed by an earlier commit) where uninitialized memory was being leaked to disk in the last byte of blocks. Also, strangely these constraints don't apply to the replay code in e2fsprogs, which will accept any tlvs in the blocks (with no bounds checks at all, but that is a separate issue...). Given that this all seems to be a bug, let's fix it by just filling blocks with tlv entries in the natural way. Note that old kernels will be unable to replay fast-commit journals created by kernels that have this commit. Fixes: aa75f4d3daae ("ext4: main fast-commit commit path") Cc: <stable@vger.kernel.org> # v5.10+ Signed-off-by: Eric Biggers <ebiggers@google.com> Link: https://lore.kernel.org/r/20221106224841.279231-7-ebiggers@kernel.org Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-11-07 06:48:40 +08:00
tl.fc_len = cpu_to_le16(remaining);
memcpy(dst, &tl, EXT4_FC_TAG_BASE_LEN);
memset(dst + EXT4_FC_TAG_BASE_LEN, 0, remaining);
*crc = ext4_chksum(sbi, *crc, sbi->s_fc_bh->b_data, bsize);
ext4_fc_submit_bh(sb, false);
ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
if (ret)
return NULL;
sbi->s_fc_bh = bh;
ext4: fix off-by-one errors in fast-commit block filling Due to several different off-by-one errors, or perhaps due to a late change in design that wasn't fully reflected in the code that was actually merged, there are several very strange constraints on how fast-commit blocks are filled with tlv entries: - tlvs must start at least 10 bytes before the end of the block, even though the minimum tlv length is 8. Otherwise, the replay code will ignore them. (BUG: ext4_fc_reserve_space() could violate this requirement if called with a len of blocksize - 9 or blocksize - 8. Fortunately, this doesn't seem to happen currently.) - tlvs must end at least 1 byte before the end of the block. Otherwise the replay code will consider them to be invalid. This quirk contributed to a bug (fixed by an earlier commit) where uninitialized memory was being leaked to disk in the last byte of blocks. Also, strangely these constraints don't apply to the replay code in e2fsprogs, which will accept any tlvs in the blocks (with no bounds checks at all, but that is a separate issue...). Given that this all seems to be a bug, let's fix it by just filling blocks with tlv entries in the natural way. Note that old kernels will be unable to replay fast-commit journals created by kernels that have this commit. Fixes: aa75f4d3daae ("ext4: main fast-commit commit path") Cc: <stable@vger.kernel.org> # v5.10+ Signed-off-by: Eric Biggers <ebiggers@google.com> Link: https://lore.kernel.org/r/20221106224841.279231-7-ebiggers@kernel.org Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-11-07 06:48:40 +08:00
sbi->s_fc_bytes += bsize - off + len;
return sbi->s_fc_bh->b_data;
}
/*
* Complete a fast commit by writing tail tag.
*
* Writing tail tag marks the end of a fast commit. In order to guarantee
* atomicity, after writing tail tag, even if there's space remaining
* in the block, next commit shouldn't use it. That's why tail tag
* has the length as that of the remaining space on the block.
*/
static int ext4_fc_write_tail(struct super_block *sb, u32 crc)
{
struct ext4_sb_info *sbi = EXT4_SB(sb);
struct ext4_fc_tl tl;
struct ext4_fc_tail tail;
int off, bsize = sbi->s_journal->j_blocksize;
u8 *dst;
/*
* ext4_fc_reserve_space takes care of allocating an extra block if
* there's no enough space on this block for accommodating this tail.
*/
dst = ext4_fc_reserve_space(sb, EXT4_FC_TAG_BASE_LEN + sizeof(tail), &crc);
if (!dst)
return -ENOSPC;
off = sbi->s_fc_bytes % bsize;
tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_TAIL);
ext4: fix off-by-one errors in fast-commit block filling Due to several different off-by-one errors, or perhaps due to a late change in design that wasn't fully reflected in the code that was actually merged, there are several very strange constraints on how fast-commit blocks are filled with tlv entries: - tlvs must start at least 10 bytes before the end of the block, even though the minimum tlv length is 8. Otherwise, the replay code will ignore them. (BUG: ext4_fc_reserve_space() could violate this requirement if called with a len of blocksize - 9 or blocksize - 8. Fortunately, this doesn't seem to happen currently.) - tlvs must end at least 1 byte before the end of the block. Otherwise the replay code will consider them to be invalid. This quirk contributed to a bug (fixed by an earlier commit) where uninitialized memory was being leaked to disk in the last byte of blocks. Also, strangely these constraints don't apply to the replay code in e2fsprogs, which will accept any tlvs in the blocks (with no bounds checks at all, but that is a separate issue...). Given that this all seems to be a bug, let's fix it by just filling blocks with tlv entries in the natural way. Note that old kernels will be unable to replay fast-commit journals created by kernels that have this commit. Fixes: aa75f4d3daae ("ext4: main fast-commit commit path") Cc: <stable@vger.kernel.org> # v5.10+ Signed-off-by: Eric Biggers <ebiggers@google.com> Link: https://lore.kernel.org/r/20221106224841.279231-7-ebiggers@kernel.org Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-11-07 06:48:40 +08:00
tl.fc_len = cpu_to_le16(bsize - off + sizeof(struct ext4_fc_tail));
sbi->s_fc_bytes = round_up(sbi->s_fc_bytes, bsize);
memcpy(dst, &tl, EXT4_FC_TAG_BASE_LEN);
dst += EXT4_FC_TAG_BASE_LEN;
tail.fc_tid = cpu_to_le32(sbi->s_journal->j_running_transaction->t_tid);
memcpy(dst, &tail.fc_tid, sizeof(tail.fc_tid));
dst += sizeof(tail.fc_tid);
crc = ext4_chksum(sbi, crc, sbi->s_fc_bh->b_data,
dst - (u8 *)sbi->s_fc_bh->b_data);
tail.fc_crc = cpu_to_le32(crc);
memcpy(dst, &tail.fc_crc, sizeof(tail.fc_crc));
dst += sizeof(tail.fc_crc);
memset(dst, 0, bsize - off); /* Don't leak uninitialized memory. */
ext4_fc_submit_bh(sb, true);
return 0;
}
/*
* Adds tag, length, value and updates CRC. Returns true if tlv was added.
* Returns false if there's not enough space.
*/
static bool ext4_fc_add_tlv(struct super_block *sb, u16 tag, u16 len, u8 *val,
u32 *crc)
{
struct ext4_fc_tl tl;
u8 *dst;
dst = ext4_fc_reserve_space(sb, EXT4_FC_TAG_BASE_LEN + len, crc);
if (!dst)
return false;
tl.fc_tag = cpu_to_le16(tag);
tl.fc_len = cpu_to_le16(len);
memcpy(dst, &tl, EXT4_FC_TAG_BASE_LEN);
memcpy(dst + EXT4_FC_TAG_BASE_LEN, val, len);
return true;
}
/* Same as above, but adds dentry tlv. */
static bool ext4_fc_add_dentry_tlv(struct super_block *sb, u32 *crc,
struct ext4_fc_dentry_update *fc_dentry)
{
struct ext4_fc_dentry_info fcd;
struct ext4_fc_tl tl;
int dlen = fc_dentry->fcd_name.len;
u8 *dst = ext4_fc_reserve_space(sb,
EXT4_FC_TAG_BASE_LEN + sizeof(fcd) + dlen, crc);
if (!dst)
return false;
fcd.fc_parent_ino = cpu_to_le32(fc_dentry->fcd_parent);
fcd.fc_ino = cpu_to_le32(fc_dentry->fcd_ino);
tl.fc_tag = cpu_to_le16(fc_dentry->fcd_op);
tl.fc_len = cpu_to_le16(sizeof(fcd) + dlen);
memcpy(dst, &tl, EXT4_FC_TAG_BASE_LEN);
dst += EXT4_FC_TAG_BASE_LEN;
memcpy(dst, &fcd, sizeof(fcd));
dst += sizeof(fcd);
memcpy(dst, fc_dentry->fcd_name.name, dlen);
return true;
}
/*
* Writes inode in the fast commit space under TLV with tag @tag.
* Returns 0 on success, error on failure.
*/
static int ext4_fc_write_inode(struct inode *inode, u32 *crc)
{
struct ext4_inode_info *ei = EXT4_I(inode);
int inode_len = EXT4_GOOD_OLD_INODE_SIZE;
int ret;
struct ext4_iloc iloc;
struct ext4_fc_inode fc_inode;
struct ext4_fc_tl tl;
u8 *dst;
ret = ext4_get_inode_loc(inode, &iloc);
if (ret)
return ret;
if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))
inode_len = EXT4_INODE_SIZE(inode->i_sb);
else if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE)
inode_len += ei->i_extra_isize;
fc_inode.fc_ino = cpu_to_le32(inode->i_ino);
tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_INODE);
tl.fc_len = cpu_to_le16(inode_len + sizeof(fc_inode.fc_ino));
ret = -ECANCELED;
dst = ext4_fc_reserve_space(inode->i_sb,
EXT4_FC_TAG_BASE_LEN + inode_len + sizeof(fc_inode.fc_ino), crc);
if (!dst)
goto err;
memcpy(dst, &tl, EXT4_FC_TAG_BASE_LEN);
dst += EXT4_FC_TAG_BASE_LEN;
memcpy(dst, &fc_inode, sizeof(fc_inode));
dst += sizeof(fc_inode);
memcpy(dst, (u8 *)ext4_raw_inode(&iloc), inode_len);
ret = 0;
err:
brelse(iloc.bh);
return ret;
}
/*
* Writes updated data ranges for the inode in question. Updates CRC.
* Returns 0 on success, error otherwise.
*/
static int ext4_fc_write_inode_data(struct inode *inode, u32 *crc)
{
ext4_lblk_t old_blk_size, cur_lblk_off, new_blk_size;
struct ext4_inode_info *ei = EXT4_I(inode);
struct ext4_map_blocks map;
struct ext4_fc_add_range fc_ext;
struct ext4_fc_del_range lrange;
struct ext4_extent *ex;
int ret;
mutex_lock(&ei->i_fc_lock);
if (ei->i_fc_lblk_len == 0) {
mutex_unlock(&ei->i_fc_lock);
return 0;
}
old_blk_size = ei->i_fc_lblk_start;
new_blk_size = ei->i_fc_lblk_start + ei->i_fc_lblk_len - 1;
ei->i_fc_lblk_len = 0;
mutex_unlock(&ei->i_fc_lock);
cur_lblk_off = old_blk_size;
ext4_debug("will try writing %d to %d for inode %ld\n",
cur_lblk_off, new_blk_size, inode->i_ino);
while (cur_lblk_off <= new_blk_size) {
map.m_lblk = cur_lblk_off;
map.m_len = new_blk_size - cur_lblk_off + 1;
ret = ext4_map_blocks(NULL, inode, &map, 0);
if (ret < 0)
return -ECANCELED;
if (map.m_len == 0) {
cur_lblk_off++;
continue;
}
if (ret == 0) {
lrange.fc_ino = cpu_to_le32(inode->i_ino);
lrange.fc_lblk = cpu_to_le32(map.m_lblk);
lrange.fc_len = cpu_to_le32(map.m_len);
if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_DEL_RANGE,
sizeof(lrange), (u8 *)&lrange, crc))
return -ENOSPC;
} else {
unsigned int max = (map.m_flags & EXT4_MAP_UNWRITTEN) ?
EXT_UNWRITTEN_MAX_LEN : EXT_INIT_MAX_LEN;
/* Limit the number of blocks in one extent */
map.m_len = min(max, map.m_len);
fc_ext.fc_ino = cpu_to_le32(inode->i_ino);
ex = (struct ext4_extent *)&fc_ext.fc_ex;
ex->ee_block = cpu_to_le32(map.m_lblk);
ex->ee_len = cpu_to_le16(map.m_len);
ext4_ext_store_pblock(ex, map.m_pblk);
if (map.m_flags & EXT4_MAP_UNWRITTEN)
ext4_ext_mark_unwritten(ex);
else
ext4_ext_mark_initialized(ex);
if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_ADD_RANGE,
sizeof(fc_ext), (u8 *)&fc_ext, crc))
return -ENOSPC;
}
cur_lblk_off += map.m_len;
}
return 0;
}
/* Submit data for all the fast commit inodes */
static int ext4_fc_submit_inode_data_all(journal_t *journal)
{
struct super_block *sb = journal->j_private;
struct ext4_sb_info *sbi = EXT4_SB(sb);
struct ext4_inode_info *ei;
int ret = 0;
spin_lock(&sbi->s_fc_lock);
list_for_each_entry(ei, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
ext4_set_inode_state(&ei->vfs_inode, EXT4_STATE_FC_COMMITTING);
while (atomic_read(&ei->i_fc_updates)) {
DEFINE_WAIT(wait);
prepare_to_wait(&ei->i_fc_wait, &wait,
TASK_UNINTERRUPTIBLE);
if (atomic_read(&ei->i_fc_updates)) {
spin_unlock(&sbi->s_fc_lock);
schedule();
spin_lock(&sbi->s_fc_lock);
}
finish_wait(&ei->i_fc_wait, &wait);
}
spin_unlock(&sbi->s_fc_lock);
ret = jbd2_submit_inode_data(journal, ei->jinode);
if (ret)
return ret;
spin_lock(&sbi->s_fc_lock);
}
spin_unlock(&sbi->s_fc_lock);
return ret;
}
/* Wait for completion of data for all the fast commit inodes */
static int ext4_fc_wait_inode_data_all(journal_t *journal)
{
struct super_block *sb = journal->j_private;
struct ext4_sb_info *sbi = EXT4_SB(sb);
struct ext4_inode_info *pos, *n;
int ret = 0;
spin_lock(&sbi->s_fc_lock);
list_for_each_entry_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
if (!ext4_test_inode_state(&pos->vfs_inode,
EXT4_STATE_FC_COMMITTING))
continue;
spin_unlock(&sbi->s_fc_lock);
ret = jbd2_wait_inode_data(journal, pos->jinode);
if (ret)
return ret;
spin_lock(&sbi->s_fc_lock);
}
spin_unlock(&sbi->s_fc_lock);
return 0;
}
/* Commit all the directory entry updates */
static int ext4_fc_commit_dentry_updates(journal_t *journal, u32 *crc)
__acquires(&sbi->s_fc_lock)
__releases(&sbi->s_fc_lock)
{
struct super_block *sb = journal->j_private;
struct ext4_sb_info *sbi = EXT4_SB(sb);
struct ext4_fc_dentry_update *fc_dentry, *fc_dentry_n;
struct inode *inode;
ext4: improve fast_commit performance and scalability Currently ext4_fc_commit_dentry_updates() is of quadratic time complexity, which is causing performance bottlenecks with high threads/file/dir count with fs_mark. This patch makes commit dentry updates (and hence ext4_fc_commit()) path to linear time complexity. Hence improves the performance of workloads which does fsync on multiple threads/open files one-by-one. Absolute numbers in avg file creates per sec (from fs_mark in 1K order) ======================================================================= no. Order without-patch(K) with-patch(K) Diff(%) 1 1 16.90 17.51 +3.60 2 2,2 32.08 31.80 -0.87 3 3,3 53.97 55.01 +1.92 4 4,4 78.94 76.90 -2.58 5 5,5 95.82 95.37 -0.46 6 6,6 87.92 103.38 +17.58 7 6,10 0.73 126.13 +17178.08 8 6,14 2.33 143.19 +6045.49 workload type ============== For e.g. 7th row order of 6,10 (2^6 == 64 && 2^10 == 1024) echo /run/riteshh/mnt/{1..64} |sed -E 's/[[:space:]]+/ -d /g' \ | xargs -I {} bash -c "sudo fs_mark -L 100 -D 1024 -n 1024 -s0 -S5 -d {}" Perf profile (w/o patches) ============================= 87.15% [kernel] [k] ext4_fc_commit --> Heavy contention/bottleneck 1.98% [kernel] [k] perf_event_interrupt 0.96% [kernel] [k] power_pmu_enable 0.91% [kernel] [k] update_sd_lb_stats.constprop.0 0.67% [kernel] [k] ktime_get Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com> Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com> Link: https://lore.kernel.org/r/930f35d4fd5f83e2673c868781d9ebf15e91bf4e.1645426817.git.riteshh@linux.ibm.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-02-21 15:56:15 +08:00
struct ext4_inode_info *ei;
int ret;
if (list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN]))
return 0;
list_for_each_entry_safe(fc_dentry, fc_dentry_n,
&sbi->s_fc_dentry_q[FC_Q_MAIN], fcd_list) {
if (fc_dentry->fcd_op != EXT4_FC_TAG_CREAT) {
spin_unlock(&sbi->s_fc_lock);
if (!ext4_fc_add_dentry_tlv(sb, crc, fc_dentry)) {
ret = -ENOSPC;
goto lock_and_exit;
}
spin_lock(&sbi->s_fc_lock);
continue;
}
/*
ext4: improve fast_commit performance and scalability Currently ext4_fc_commit_dentry_updates() is of quadratic time complexity, which is causing performance bottlenecks with high threads/file/dir count with fs_mark. This patch makes commit dentry updates (and hence ext4_fc_commit()) path to linear time complexity. Hence improves the performance of workloads which does fsync on multiple threads/open files one-by-one. Absolute numbers in avg file creates per sec (from fs_mark in 1K order) ======================================================================= no. Order without-patch(K) with-patch(K) Diff(%) 1 1 16.90 17.51 +3.60 2 2,2 32.08 31.80 -0.87 3 3,3 53.97 55.01 +1.92 4 4,4 78.94 76.90 -2.58 5 5,5 95.82 95.37 -0.46 6 6,6 87.92 103.38 +17.58 7 6,10 0.73 126.13 +17178.08 8 6,14 2.33 143.19 +6045.49 workload type ============== For e.g. 7th row order of 6,10 (2^6 == 64 && 2^10 == 1024) echo /run/riteshh/mnt/{1..64} |sed -E 's/[[:space:]]+/ -d /g' \ | xargs -I {} bash -c "sudo fs_mark -L 100 -D 1024 -n 1024 -s0 -S5 -d {}" Perf profile (w/o patches) ============================= 87.15% [kernel] [k] ext4_fc_commit --> Heavy contention/bottleneck 1.98% [kernel] [k] perf_event_interrupt 0.96% [kernel] [k] power_pmu_enable 0.91% [kernel] [k] update_sd_lb_stats.constprop.0 0.67% [kernel] [k] ktime_get Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com> Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com> Link: https://lore.kernel.org/r/930f35d4fd5f83e2673c868781d9ebf15e91bf4e.1645426817.git.riteshh@linux.ibm.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-02-21 15:56:15 +08:00
* With fcd_dilist we need not loop in sbi->s_fc_q to get the
* corresponding inode pointer
*/
ext4: improve fast_commit performance and scalability Currently ext4_fc_commit_dentry_updates() is of quadratic time complexity, which is causing performance bottlenecks with high threads/file/dir count with fs_mark. This patch makes commit dentry updates (and hence ext4_fc_commit()) path to linear time complexity. Hence improves the performance of workloads which does fsync on multiple threads/open files one-by-one. Absolute numbers in avg file creates per sec (from fs_mark in 1K order) ======================================================================= no. Order without-patch(K) with-patch(K) Diff(%) 1 1 16.90 17.51 +3.60 2 2,2 32.08 31.80 -0.87 3 3,3 53.97 55.01 +1.92 4 4,4 78.94 76.90 -2.58 5 5,5 95.82 95.37 -0.46 6 6,6 87.92 103.38 +17.58 7 6,10 0.73 126.13 +17178.08 8 6,14 2.33 143.19 +6045.49 workload type ============== For e.g. 7th row order of 6,10 (2^6 == 64 && 2^10 == 1024) echo /run/riteshh/mnt/{1..64} |sed -E 's/[[:space:]]+/ -d /g' \ | xargs -I {} bash -c "sudo fs_mark -L 100 -D 1024 -n 1024 -s0 -S5 -d {}" Perf profile (w/o patches) ============================= 87.15% [kernel] [k] ext4_fc_commit --> Heavy contention/bottleneck 1.98% [kernel] [k] perf_event_interrupt 0.96% [kernel] [k] power_pmu_enable 0.91% [kernel] [k] update_sd_lb_stats.constprop.0 0.67% [kernel] [k] ktime_get Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com> Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com> Link: https://lore.kernel.org/r/930f35d4fd5f83e2673c868781d9ebf15e91bf4e.1645426817.git.riteshh@linux.ibm.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-02-21 15:56:15 +08:00
WARN_ON(list_empty(&fc_dentry->fcd_dilist));
ei = list_first_entry(&fc_dentry->fcd_dilist,
struct ext4_inode_info, i_fc_dilist);
inode = &ei->vfs_inode;
WARN_ON(inode->i_ino != fc_dentry->fcd_ino);
spin_unlock(&sbi->s_fc_lock);
/*
* We first write the inode and then the create dirent. This
* allows the recovery code to create an unnamed inode first
* and then link it to a directory entry. This allows us
* to use namei.c routines almost as is and simplifies
* the recovery code.
*/
ret = ext4_fc_write_inode(inode, crc);
if (ret)
goto lock_and_exit;
ret = ext4_fc_write_inode_data(inode, crc);
if (ret)
goto lock_and_exit;
if (!ext4_fc_add_dentry_tlv(sb, crc, fc_dentry)) {
ret = -ENOSPC;
goto lock_and_exit;
}
spin_lock(&sbi->s_fc_lock);
}
return 0;
lock_and_exit:
spin_lock(&sbi->s_fc_lock);
return ret;
}
static int ext4_fc_perform_commit(journal_t *journal)
{
struct super_block *sb = journal->j_private;
struct ext4_sb_info *sbi = EXT4_SB(sb);
struct ext4_inode_info *iter;
struct ext4_fc_head head;
struct inode *inode;
struct blk_plug plug;
int ret = 0;
u32 crc = 0;
ret = ext4_fc_submit_inode_data_all(journal);
if (ret)
return ret;
ret = ext4_fc_wait_inode_data_all(journal);
if (ret)
return ret;
/*
* If file system device is different from journal device, issue a cache
* flush before we start writing fast commit blocks.
*/
if (journal->j_fs_dev != journal->j_dev)
blkdev_issue_flush(journal->j_fs_dev);
blk_start_plug(&plug);
if (sbi->s_fc_bytes == 0) {
/*
* Add a head tag only if this is the first fast commit
* in this TID.
*/
head.fc_features = cpu_to_le32(EXT4_FC_SUPPORTED_FEATURES);
head.fc_tid = cpu_to_le32(
sbi->s_journal->j_running_transaction->t_tid);
if (!ext4_fc_add_tlv(sb, EXT4_FC_TAG_HEAD, sizeof(head),
(u8 *)&head, &crc)) {
ret = -ENOSPC;
goto out;
}
}
spin_lock(&sbi->s_fc_lock);
ret = ext4_fc_commit_dentry_updates(journal, &crc);
if (ret) {
spin_unlock(&sbi->s_fc_lock);
goto out;
}
list_for_each_entry(iter, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
inode = &iter->vfs_inode;
if (!ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING))
continue;
spin_unlock(&sbi->s_fc_lock);
ret = ext4_fc_write_inode_data(inode, &crc);
if (ret)
goto out;
ret = ext4_fc_write_inode(inode, &crc);
if (ret)
goto out;
spin_lock(&sbi->s_fc_lock);
}
spin_unlock(&sbi->s_fc_lock);
ret = ext4_fc_write_tail(sb, crc);
out:
blk_finish_plug(&plug);
return ret;
}
static void ext4_fc_update_stats(struct super_block *sb, int status,
u64 commit_time, int nblks, tid_t commit_tid)
{
struct ext4_fc_stats *stats = &EXT4_SB(sb)->s_fc_stats;
ext4_debug("Fast commit ended with status = %d for tid %u",
status, commit_tid);
if (status == EXT4_FC_STATUS_OK) {
stats->fc_num_commits++;
stats->fc_numblks += nblks;
if (likely(stats->s_fc_avg_commit_time))
stats->s_fc_avg_commit_time =
(commit_time +
stats->s_fc_avg_commit_time * 3) / 4;
else
stats->s_fc_avg_commit_time = commit_time;
} else if (status == EXT4_FC_STATUS_FAILED ||
status == EXT4_FC_STATUS_INELIGIBLE) {
if (status == EXT4_FC_STATUS_FAILED)
stats->fc_failed_commits++;
stats->fc_ineligible_commits++;
} else {
stats->fc_skipped_commits++;
}
trace_ext4_fc_commit_stop(sb, nblks, status, commit_tid);
}
/*
* The main commit entry point. Performs a fast commit for transaction
* commit_tid if needed. If it's not possible to perform a fast commit
* due to various reasons, we fall back to full commit. Returns 0
* on success, error otherwise.
*/
int ext4_fc_commit(journal_t *journal, tid_t commit_tid)
{
struct super_block *sb = journal->j_private;
struct ext4_sb_info *sbi = EXT4_SB(sb);
int nblks = 0, ret, bsize = journal->j_blocksize;
int subtid = atomic_read(&sbi->s_fc_subtid);
int status = EXT4_FC_STATUS_OK, fc_bufs_before = 0;
ktime_t start_time, commit_time;
if (!test_opt2(sb, JOURNAL_FAST_COMMIT))
return jbd2_complete_transaction(journal, commit_tid);
trace_ext4_fc_commit_start(sb, commit_tid);
start_time = ktime_get();
restart_fc:
ret = jbd2_fc_begin_commit(journal, commit_tid);
if (ret == -EALREADY) {
/* There was an ongoing commit, check if we need to restart */
if (atomic_read(&sbi->s_fc_subtid) <= subtid &&
commit_tid > journal->j_commit_sequence)
goto restart_fc;
ext4_fc_update_stats(sb, EXT4_FC_STATUS_SKIPPED, 0, 0,
commit_tid);
return 0;
} else if (ret) {
/*
* Commit couldn't start. Just update stats and perform a
* full commit.
*/
ext4_fc_update_stats(sb, EXT4_FC_STATUS_FAILED, 0, 0,
commit_tid);
return jbd2_complete_transaction(journal, commit_tid);
}
/*
* After establishing journal barrier via jbd2_fc_begin_commit(), check
* if we are fast commit ineligible.
*/
if (ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE)) {
status = EXT4_FC_STATUS_INELIGIBLE;
goto fallback;
}
fc_bufs_before = (sbi->s_fc_bytes + bsize - 1) / bsize;
ret = ext4_fc_perform_commit(journal);
if (ret < 0) {
status = EXT4_FC_STATUS_FAILED;
goto fallback;
}
nblks = (sbi->s_fc_bytes + bsize - 1) / bsize - fc_bufs_before;
ret = jbd2_fc_wait_bufs(journal, nblks);
if (ret < 0) {
status = EXT4_FC_STATUS_FAILED;
goto fallback;
}
atomic_inc(&sbi->s_fc_subtid);
ret = jbd2_fc_end_commit(journal);
/*
* weight the commit time higher than the average time so we
* don't react too strongly to vast changes in the commit time
*/
commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
ext4_fc_update_stats(sb, status, commit_time, nblks, commit_tid);
return ret;
fallback:
ret = jbd2_fc_end_commit_fallback(journal);
ext4_fc_update_stats(sb, status, 0, 0, commit_tid);
return ret;
}
/*
* Fast commit cleanup routine. This is called after every fast commit and
* full commit. full is true if we are called after a full commit.
*/
static void ext4_fc_cleanup(journal_t *journal, int full, tid_t tid)
{
struct super_block *sb = journal->j_private;
struct ext4_sb_info *sbi = EXT4_SB(sb);
struct ext4_inode_info *iter, *iter_n;
struct ext4_fc_dentry_update *fc_dentry;
if (full && sbi->s_fc_bh)
sbi->s_fc_bh = NULL;
trace_ext4_fc_cleanup(journal, full, tid);
jbd2_fc_release_bufs(journal);
spin_lock(&sbi->s_fc_lock);
list_for_each_entry_safe(iter, iter_n, &sbi->s_fc_q[FC_Q_MAIN],
i_fc_list) {
list_del_init(&iter->i_fc_list);
ext4_clear_inode_state(&iter->vfs_inode,
EXT4_STATE_FC_COMMITTING);
if (iter->i_sync_tid <= tid)
ext4_fc_reset_inode(&iter->vfs_inode);
/* Make sure EXT4_STATE_FC_COMMITTING bit is clear */
smp_mb();
#if (BITS_PER_LONG < 64)
wake_up_bit(&iter->i_state_flags, EXT4_STATE_FC_COMMITTING);
#else
wake_up_bit(&iter->i_flags, EXT4_STATE_FC_COMMITTING);
#endif
}
while (!list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN])) {
fc_dentry = list_first_entry(&sbi->s_fc_dentry_q[FC_Q_MAIN],
struct ext4_fc_dentry_update,
fcd_list);
list_del_init(&fc_dentry->fcd_list);
ext4: improve fast_commit performance and scalability Currently ext4_fc_commit_dentry_updates() is of quadratic time complexity, which is causing performance bottlenecks with high threads/file/dir count with fs_mark. This patch makes commit dentry updates (and hence ext4_fc_commit()) path to linear time complexity. Hence improves the performance of workloads which does fsync on multiple threads/open files one-by-one. Absolute numbers in avg file creates per sec (from fs_mark in 1K order) ======================================================================= no. Order without-patch(K) with-patch(K) Diff(%) 1 1 16.90 17.51 +3.60 2 2,2 32.08 31.80 -0.87 3 3,3 53.97 55.01 +1.92 4 4,4 78.94 76.90 -2.58 5 5,5 95.82 95.37 -0.46 6 6,6 87.92 103.38 +17.58 7 6,10 0.73 126.13 +17178.08 8 6,14 2.33 143.19 +6045.49 workload type ============== For e.g. 7th row order of 6,10 (2^6 == 64 && 2^10 == 1024) echo /run/riteshh/mnt/{1..64} |sed -E 's/[[:space:]]+/ -d /g' \ | xargs -I {} bash -c "sudo fs_mark -L 100 -D 1024 -n 1024 -s0 -S5 -d {}" Perf profile (w/o patches) ============================= 87.15% [kernel] [k] ext4_fc_commit --> Heavy contention/bottleneck 1.98% [kernel] [k] perf_event_interrupt 0.96% [kernel] [k] power_pmu_enable 0.91% [kernel] [k] update_sd_lb_stats.constprop.0 0.67% [kernel] [k] ktime_get Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com> Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com> Link: https://lore.kernel.org/r/930f35d4fd5f83e2673c868781d9ebf15e91bf4e.1645426817.git.riteshh@linux.ibm.com Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-02-21 15:56:15 +08:00
list_del_init(&fc_dentry->fcd_dilist);
spin_unlock(&sbi->s_fc_lock);
if (fc_dentry->fcd_name.name &&
fc_dentry->fcd_name.len > DNAME_INLINE_LEN)
kfree(fc_dentry->fcd_name.name);
kmem_cache_free(ext4_fc_dentry_cachep, fc_dentry);
spin_lock(&sbi->s_fc_lock);
}
list_splice_init(&sbi->s_fc_dentry_q[FC_Q_STAGING],
&sbi->s_fc_dentry_q[FC_Q_MAIN]);
list_splice_init(&sbi->s_fc_q[FC_Q_STAGING],
&sbi->s_fc_q[FC_Q_MAIN]);
if (tid >= sbi->s_fc_ineligible_tid) {
sbi->s_fc_ineligible_tid = 0;
ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
}
if (full)
sbi->s_fc_bytes = 0;
spin_unlock(&sbi->s_fc_lock);
trace_ext4_fc_stats(sb);
}
/* Ext4 Replay Path Routines */
/* Helper struct for dentry replay routines */
struct dentry_info_args {
int parent_ino, dname_len, ino, inode_len;
char *dname;
};
/* Same as struct ext4_fc_tl, but uses native endianness fields */
struct ext4_fc_tl_mem {
u16 fc_tag;
u16 fc_len;
};
static inline void tl_to_darg(struct dentry_info_args *darg,
struct ext4_fc_tl_mem *tl, u8 *val)
{
struct ext4_fc_dentry_info fcd;
memcpy(&fcd, val, sizeof(fcd));
darg->parent_ino = le32_to_cpu(fcd.fc_parent_ino);
darg->ino = le32_to_cpu(fcd.fc_ino);
darg->dname = val + offsetof(struct ext4_fc_dentry_info, fc_dname);
darg->dname_len = tl->fc_len - sizeof(struct ext4_fc_dentry_info);
}
static inline void ext4_fc_get_tl(struct ext4_fc_tl_mem *tl, u8 *val)
{
struct ext4_fc_tl tl_disk;
memcpy(&tl_disk, val, EXT4_FC_TAG_BASE_LEN);
tl->fc_len = le16_to_cpu(tl_disk.fc_len);
tl->fc_tag = le16_to_cpu(tl_disk.fc_tag);
}
/* Unlink replay function */
static int ext4_fc_replay_unlink(struct super_block *sb,
struct ext4_fc_tl_mem *tl, u8 *val)
{
struct inode *inode, *old_parent;
struct qstr entry;
struct dentry_info_args darg;
int ret = 0;
tl_to_darg(&darg, tl, val);
trace_ext4_fc_replay(sb, EXT4_FC_TAG_UNLINK, darg.ino,
darg.parent_ino, darg.dname_len);
entry.name = darg.dname;
entry.len = darg.dname_len;
inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
if (IS_ERR(inode)) {
ext4_debug("Inode %d not found", darg.ino);
return 0;
}
old_parent = ext4_iget(sb, darg.parent_ino,
EXT4_IGET_NORMAL);
if (IS_ERR(old_parent)) {
ext4_debug("Dir with inode %d not found", darg.parent_ino);
iput(inode);
return 0;
}
ret = __ext4_unlink(old_parent, &entry, inode, NULL);
/* -ENOENT ok coz it might not exist anymore. */
if (ret == -ENOENT)
ret = 0;
iput(old_parent);
iput(inode);
return ret;
}
static int ext4_fc_replay_link_internal(struct super_block *sb,
struct dentry_info_args *darg,
struct inode *inode)
{
struct inode *dir = NULL;
struct dentry *dentry_dir = NULL, *dentry_inode = NULL;
struct qstr qstr_dname = QSTR_INIT(darg->dname, darg->dname_len);
int ret = 0;
dir = ext4_iget(sb, darg->parent_ino, EXT4_IGET_NORMAL);
if (IS_ERR(dir)) {
ext4_debug("Dir with inode %d not found.", darg->parent_ino);
dir = NULL;
goto out;
}
dentry_dir = d_obtain_alias(dir);
if (IS_ERR(dentry_dir)) {
ext4_debug("Failed to obtain dentry");
dentry_dir = NULL;
goto out;
}
dentry_inode = d_alloc(dentry_dir, &qstr_dname);
if (!dentry_inode) {
ext4_debug("Inode dentry not created.");
ret = -ENOMEM;
goto out;
}
ret = __ext4_link(dir, inode, dentry_inode);
/*
* It's possible that link already existed since data blocks
* for the dir in question got persisted before we crashed OR
* we replayed this tag and crashed before the entire replay
* could complete.
*/
if (ret && ret != -EEXIST) {
ext4_debug("Failed to link\n");
goto out;
}
ret = 0;
out:
if (dentry_dir) {
d_drop(dentry_dir);
dput(dentry_dir);
} else if (dir) {
iput(dir);
}
if (dentry_inode) {
d_drop(dentry_inode);
dput(dentry_inode);
}
return ret;
}
/* Link replay function */
static int ext4_fc_replay_link(struct super_block *sb,
struct ext4_fc_tl_mem *tl, u8 *val)
{
struct inode *inode;
struct dentry_info_args darg;
int ret = 0;
tl_to_darg(&darg, tl, val);
trace_ext4_fc_replay(sb, EXT4_FC_TAG_LINK, darg.ino,
darg.parent_ino, darg.dname_len);
inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
if (IS_ERR(inode)) {
ext4_debug("Inode not found.");
return 0;
}
ret = ext4_fc_replay_link_internal(sb, &darg, inode);
iput(inode);
return ret;
}
/*
* Record all the modified inodes during replay. We use this later to setup
* block bitmaps correctly.
*/
static int ext4_fc_record_modified_inode(struct super_block *sb, int ino)
{
struct ext4_fc_replay_state *state;
int i;
state = &EXT4_SB(sb)->s_fc_replay_state;
for (i = 0; i < state->fc_modified_inodes_used; i++)
if (state->fc_modified_inodes[i] == ino)
return 0;
if (state->fc_modified_inodes_used == state->fc_modified_inodes_size) {
int *fc_modified_inodes;
fc_modified_inodes = krealloc(state->fc_modified_inodes,
sizeof(int) * (state->fc_modified_inodes_size +
EXT4_FC_REPLAY_REALLOC_INCREMENT),
GFP_KERNEL);
if (!fc_modified_inodes)
return -ENOMEM;
state->fc_modified_inodes = fc_modified_inodes;
state->fc_modified_inodes_size +=
EXT4_FC_REPLAY_REALLOC_INCREMENT;
}
state->fc_modified_inodes[state->fc_modified_inodes_used++] = ino;
return 0;
}
/*
* Inode replay function
*/
static int ext4_fc_replay_inode(struct super_block *sb,
struct ext4_fc_tl_mem *tl, u8 *val)
{
struct ext4_fc_inode fc_inode;
struct ext4_inode *raw_inode;
struct ext4_inode *raw_fc_inode;
struct inode *inode = NULL;
struct ext4_iloc iloc;
int inode_len, ino, ret, tag = tl->fc_tag;
struct ext4_extent_header *eh;
size_t off_gen = offsetof(struct ext4_inode, i_generation);
memcpy(&fc_inode, val, sizeof(fc_inode));
ino = le32_to_cpu(fc_inode.fc_ino);
trace_ext4_fc_replay(sb, tag, ino, 0, 0);
inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
if (!IS_ERR(inode)) {
ext4_ext_clear_bb(inode);
iput(inode);
}
inode = NULL;
ret = ext4_fc_record_modified_inode(sb, ino);
if (ret)
goto out;
raw_fc_inode = (struct ext4_inode *)
(val + offsetof(struct ext4_fc_inode, fc_raw_inode));
ret = ext4_get_fc_inode_loc(sb, ino, &iloc);
if (ret)
goto out;
inode_len = tl->fc_len - sizeof(struct ext4_fc_inode);
raw_inode = ext4_raw_inode(&iloc);
memcpy(raw_inode, raw_fc_inode, offsetof(struct ext4_inode, i_block));
memcpy((u8 *)raw_inode + off_gen, (u8 *)raw_fc_inode + off_gen,
inode_len - off_gen);
if (le32_to_cpu(raw_inode->i_flags) & EXT4_EXTENTS_FL) {
eh = (struct ext4_extent_header *)(&raw_inode->i_block[0]);
if (eh->eh_magic != EXT4_EXT_MAGIC) {
memset(eh, 0, sizeof(*eh));
eh->eh_magic = EXT4_EXT_MAGIC;
eh->eh_max = cpu_to_le16(
(sizeof(raw_inode->i_block) -
sizeof(struct ext4_extent_header))
/ sizeof(struct ext4_extent));
}
} else if (le32_to_cpu(raw_inode->i_flags) & EXT4_INLINE_DATA_FL) {
memcpy(raw_inode->i_block, raw_fc_inode->i_block,
sizeof(raw_inode->i_block));
}
/* Immediately update the inode on disk. */
ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
if (ret)
goto out;
ret = sync_dirty_buffer(iloc.bh);
if (ret)
goto out;
ret = ext4_mark_inode_used(sb, ino);
if (ret)
goto out;
/* Given that we just wrote the inode on disk, this SHOULD succeed. */
inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
if (IS_ERR(inode)) {
ext4_debug("Inode not found.");
return -EFSCORRUPTED;
}
/*
* Our allocator could have made different decisions than before
* crashing. This should be fixed but until then, we calculate
* the number of blocks the inode.
*/
if (!ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))
ext4_ext_replay_set_iblocks(inode);
inode->i_generation = le32_to_cpu(ext4_raw_inode(&iloc)->i_generation);
ext4_reset_inode_seed(inode);
ext4_inode_csum_set(inode, ext4_raw_inode(&iloc), EXT4_I(inode));
ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
sync_dirty_buffer(iloc.bh);
brelse(iloc.bh);
out:
iput(inode);
if (!ret)
blkdev_issue_flush(sb->s_bdev);
return 0;
}
/*
* Dentry create replay function.
*
* EXT4_FC_TAG_CREAT is preceded by EXT4_FC_TAG_INODE_FULL. Which means, the
* inode for which we are trying to create a dentry here, should already have
* been replayed before we start here.
*/
static int ext4_fc_replay_create(struct super_block *sb,
struct ext4_fc_tl_mem *tl, u8 *val)
{
int ret = 0;
struct inode *inode = NULL;
struct inode *dir = NULL;
struct dentry_info_args darg;
tl_to_darg(&darg, tl, val);
trace_ext4_fc_replay(sb, EXT4_FC_TAG_CREAT, darg.ino,
darg.parent_ino, darg.dname_len);
/* This takes care of update group descriptor and other metadata */
ret = ext4_mark_inode_used(sb, darg.ino);
if (ret)
goto out;
inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
if (IS_ERR(inode)) {
ext4_debug("inode %d not found.", darg.ino);
inode = NULL;
ret = -EINVAL;
goto out;
}
if (S_ISDIR(inode->i_mode)) {
/*
* If we are creating a directory, we need to make sure that the
* dot and dot dot dirents are setup properly.
*/
dir = ext4_iget(sb, darg.parent_ino, EXT4_IGET_NORMAL);
if (IS_ERR(dir)) {
ext4_debug("Dir %d not found.", darg.ino);
goto out;
}
ret = ext4_init_new_dir(NULL, dir, inode);
iput(dir);
if (ret) {
ret = 0;
goto out;
}
}
ret = ext4_fc_replay_link_internal(sb, &darg, inode);
if (ret)
goto out;
set_nlink(inode, 1);
ext4_mark_inode_dirty(NULL, inode);
out:
iput(inode);
return ret;
}
/*
* Record physical disk regions which are in use as per fast commit area,
* and used by inodes during replay phase. Our simple replay phase
* allocator excludes these regions from allocation.
*/
int ext4_fc_record_regions(struct super_block *sb, int ino,
ext4_lblk_t lblk, ext4_fsblk_t pblk, int len, int replay)
{
struct ext4_fc_replay_state *state;
struct ext4_fc_alloc_region *region;
state = &EXT4_SB(sb)->s_fc_replay_state;
/*
* during replay phase, the fc_regions_valid may not same as
* fc_regions_used, update it when do new additions.
*/
if (replay && state->fc_regions_used != state->fc_regions_valid)
state->fc_regions_used = state->fc_regions_valid;
if (state->fc_regions_used == state->fc_regions_size) {
struct ext4_fc_alloc_region *fc_regions;
fc_regions = krealloc(state->fc_regions,
sizeof(struct ext4_fc_alloc_region) *
(state->fc_regions_size +
EXT4_FC_REPLAY_REALLOC_INCREMENT),
GFP_KERNEL);
if (!fc_regions)
return -ENOMEM;
state->fc_regions_size +=
EXT4_FC_REPLAY_REALLOC_INCREMENT;
state->fc_regions = fc_regions;
}
region = &state->fc_regions[state->fc_regions_used++];
region->ino = ino;
region->lblk = lblk;
region->pblk = pblk;
region->len = len;
if (replay)
state->fc_regions_valid++;
return 0;
}
/* Replay add range tag */
static int ext4_fc_replay_add_range(struct super_block *sb,
struct ext4_fc_tl_mem *tl, u8 *val)
{
struct ext4_fc_add_range fc_add_ex;
struct ext4_extent newex, *ex;
struct inode *inode;
ext4_lblk_t start, cur;
int remaining, len;
ext4_fsblk_t start_pblk;
struct ext4_map_blocks map;
struct ext4_ext_path *path = NULL;
int ret;
memcpy(&fc_add_ex, val, sizeof(fc_add_ex));
ex = (struct ext4_extent *)&fc_add_ex.fc_ex;
trace_ext4_fc_replay(sb, EXT4_FC_TAG_ADD_RANGE,
le32_to_cpu(fc_add_ex.fc_ino), le32_to_cpu(ex->ee_block),
ext4_ext_get_actual_len(ex));
inode = ext4_iget(sb, le32_to_cpu(fc_add_ex.fc_ino), EXT4_IGET_NORMAL);
if (IS_ERR(inode)) {
ext4_debug("Inode not found.");
return 0;
}
ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
if (ret)
goto out;
start = le32_to_cpu(ex->ee_block);
start_pblk = ext4_ext_pblock(ex);
len = ext4_ext_get_actual_len(ex);
cur = start;
remaining = len;
ext4_debug("ADD_RANGE, lblk %d, pblk %lld, len %d, unwritten %d, inode %ld\n",
start, start_pblk, len, ext4_ext_is_unwritten(ex),
inode->i_ino);
while (remaining > 0) {
map.m_lblk = cur;
map.m_len = remaining;
map.m_pblk = 0;
ret = ext4_map_blocks(NULL, inode, &map, 0);
if (ret < 0)
goto out;
if (ret == 0) {
/* Range is not mapped */
path = ext4_find_extent(inode, cur, NULL, 0);
if (IS_ERR(path))
goto out;
memset(&newex, 0, sizeof(newex));
newex.ee_block = cpu_to_le32(cur);
ext4_ext_store_pblock(
&newex, start_pblk + cur - start);
newex.ee_len = cpu_to_le16(map.m_len);
if (ext4_ext_is_unwritten(ex))
ext4_ext_mark_unwritten(&newex);
down_write(&EXT4_I(inode)->i_data_sem);
ret = ext4_ext_insert_extent(
NULL, inode, &path, &newex, 0);
up_write((&EXT4_I(inode)->i_data_sem));
ext4_free_ext_path(path);
if (ret)
goto out;
goto next;
}
if (start_pblk + cur - start != map.m_pblk) {
/*
* Logical to physical mapping changed. This can happen
* if this range was removed and then reallocated to
* map to new physical blocks during a fast commit.
*/
ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
ext4_ext_is_unwritten(ex),
start_pblk + cur - start);
if (ret)
goto out;
/*
* Mark the old blocks as free since they aren't used
* anymore. We maintain an array of all the modified
* inodes. In case these blocks are still used at either
* a different logical range in the same inode or in
* some different inode, we will mark them as allocated
* at the end of the FC replay using our array of
* modified inodes.
*/
ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
goto next;
}
/* Range is mapped and needs a state change */
ext4_debug("Converting from %ld to %d %lld",
map.m_flags & EXT4_MAP_UNWRITTEN,
ext4_ext_is_unwritten(ex), map.m_pblk);
ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
ext4_ext_is_unwritten(ex), map.m_pblk);
if (ret)
goto out;
/*
* We may have split the extent tree while toggling the state.
* Try to shrink the extent tree now.
*/
ext4_ext_replay_shrink_inode(inode, start + len);
next:
cur += map.m_len;
remaining -= map.m_len;
}
ext4_ext_replay_shrink_inode(inode, i_size_read(inode) >>
sb->s_blocksize_bits);
out:
iput(inode);
return 0;
}
/* Replay DEL_RANGE tag */
static int
ext4_fc_replay_del_range(struct super_block *sb,
struct ext4_fc_tl_mem *tl, u8 *val)
{
struct inode *inode;
struct ext4_fc_del_range lrange;
struct ext4_map_blocks map;
ext4_lblk_t cur, remaining;
int ret;
memcpy(&lrange, val, sizeof(lrange));
cur = le32_to_cpu(lrange.fc_lblk);
remaining = le32_to_cpu(lrange.fc_len);
trace_ext4_fc_replay(sb, EXT4_FC_TAG_DEL_RANGE,
le32_to_cpu(lrange.fc_ino), cur, remaining);
inode = ext4_iget(sb, le32_to_cpu(lrange.fc_ino), EXT4_IGET_NORMAL);
if (IS_ERR(inode)) {
ext4_debug("Inode %d not found", le32_to_cpu(lrange.fc_ino));
return 0;
}
ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
if (ret)
goto out;
ext4_debug("DEL_RANGE, inode %ld, lblk %d, len %d\n",
inode->i_ino, le32_to_cpu(lrange.fc_lblk),
le32_to_cpu(lrange.fc_len));
while (remaining > 0) {
map.m_lblk = cur;
map.m_len = remaining;
ret = ext4_map_blocks(NULL, inode, &map, 0);
if (ret < 0)
goto out;
if (ret > 0) {
remaining -= ret;
cur += ret;
ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
} else {
remaining -= map.m_len;
cur += map.m_len;
}
}
down_write(&EXT4_I(inode)->i_data_sem);
ret = ext4_ext_remove_space(inode, le32_to_cpu(lrange.fc_lblk),
le32_to_cpu(lrange.fc_lblk) +
le32_to_cpu(lrange.fc_len) - 1);
up_write(&EXT4_I(inode)->i_data_sem);
if (ret)
goto out;
ext4_ext_replay_shrink_inode(inode,
i_size_read(inode) >> sb->s_blocksize_bits);
ext4_mark_inode_dirty(NULL, inode);
out:
iput(inode);
return 0;
}
static void ext4_fc_set_bitmaps_and_counters(struct super_block *sb)
{
struct ext4_fc_replay_state *state;
struct inode *inode;
struct ext4_ext_path *path = NULL;
struct ext4_map_blocks map;
int i, ret, j;
ext4_lblk_t cur, end;
state = &EXT4_SB(sb)->s_fc_replay_state;
for (i = 0; i < state->fc_modified_inodes_used; i++) {
inode = ext4_iget(sb, state->fc_modified_inodes[i],
EXT4_IGET_NORMAL);
if (IS_ERR(inode)) {
ext4_debug("Inode %d not found.",
state->fc_modified_inodes[i]);
continue;
}
cur = 0;
end = EXT_MAX_BLOCKS;
if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA)) {
iput(inode);
continue;
}
while (cur < end) {
map.m_lblk = cur;
map.m_len = end - cur;
ret = ext4_map_blocks(NULL, inode, &map, 0);
if (ret < 0)
break;
if (ret > 0) {
path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
if (!IS_ERR(path)) {
for (j = 0; j < path->p_depth; j++)
ext4_mb_mark_bb(inode->i_sb,
path[j].p_block, 1, 1);
ext4_free_ext_path(path);
}
cur += ret;
ext4_mb_mark_bb(inode->i_sb, map.m_pblk,
map.m_len, 1);
} else {
cur = cur + (map.m_len ? map.m_len : 1);
}
}
iput(inode);
}
}
/*
* Check if block is in excluded regions for block allocation. The simple
* allocator that runs during replay phase is calls this function to see
* if it is okay to use a block.
*/
bool ext4_fc_replay_check_excluded(struct super_block *sb, ext4_fsblk_t blk)
{
int i;
struct ext4_fc_replay_state *state;
state = &EXT4_SB(sb)->s_fc_replay_state;
for (i = 0; i < state->fc_regions_valid; i++) {
if (state->fc_regions[i].ino == 0 ||
state->fc_regions[i].len == 0)
continue;
if (in_range(blk, state->fc_regions[i].pblk,
state->fc_regions[i].len))
return true;
}
return false;
}
/* Cleanup function called after replay */
void ext4_fc_replay_cleanup(struct super_block *sb)
{
struct ext4_sb_info *sbi = EXT4_SB(sb);
sbi->s_mount_state &= ~EXT4_FC_REPLAY;
kfree(sbi->s_fc_replay_state.fc_regions);
kfree(sbi->s_fc_replay_state.fc_modified_inodes);
}
static bool ext4_fc_value_len_isvalid(struct ext4_sb_info *sbi,
int tag, int len)
{
switch (tag) {
case EXT4_FC_TAG_ADD_RANGE:
return len == sizeof(struct ext4_fc_add_range);
case EXT4_FC_TAG_DEL_RANGE:
return len == sizeof(struct ext4_fc_del_range);
case EXT4_FC_TAG_CREAT:
case EXT4_FC_TAG_LINK:
case EXT4_FC_TAG_UNLINK:
len -= sizeof(struct ext4_fc_dentry_info);
return len >= 1 && len <= EXT4_NAME_LEN;
case EXT4_FC_TAG_INODE:
len -= sizeof(struct ext4_fc_inode);
return len >= EXT4_GOOD_OLD_INODE_SIZE &&
len <= sbi->s_inode_size;
case EXT4_FC_TAG_PAD:
return true; /* padding can have any length */
case EXT4_FC_TAG_TAIL:
return len >= sizeof(struct ext4_fc_tail);
case EXT4_FC_TAG_HEAD:
return len == sizeof(struct ext4_fc_head);
}
return false;
}
/*
* Recovery Scan phase handler
*
* This function is called during the scan phase and is responsible
* for doing following things:
* - Make sure the fast commit area has valid tags for replay
* - Count number of tags that need to be replayed by the replay handler
* - Verify CRC
* - Create a list of excluded blocks for allocation during replay phase
*
* This function returns JBD2_FC_REPLAY_CONTINUE to indicate that SCAN is
* incomplete and JBD2 should send more blocks. It returns JBD2_FC_REPLAY_STOP
* to indicate that scan has finished and JBD2 can now start replay phase.
* It returns a negative error to indicate that there was an error. At the end
* of a successful scan phase, sbi->s_fc_replay_state.fc_replay_num_tags is set
* to indicate the number of tags that need to replayed during the replay phase.
*/
static int ext4_fc_replay_scan(journal_t *journal,
struct buffer_head *bh, int off,
tid_t expected_tid)
{
struct super_block *sb = journal->j_private;
struct ext4_sb_info *sbi = EXT4_SB(sb);
struct ext4_fc_replay_state *state;
int ret = JBD2_FC_REPLAY_CONTINUE;
struct ext4_fc_add_range ext;
struct ext4_fc_tl_mem tl;
struct ext4_fc_tail tail;
__u8 *start, *end, *cur, *val;
struct ext4_fc_head head;
struct ext4_extent *ex;
state = &sbi->s_fc_replay_state;
start = (u8 *)bh->b_data;
ext4: fix off-by-one errors in fast-commit block filling Due to several different off-by-one errors, or perhaps due to a late change in design that wasn't fully reflected in the code that was actually merged, there are several very strange constraints on how fast-commit blocks are filled with tlv entries: - tlvs must start at least 10 bytes before the end of the block, even though the minimum tlv length is 8. Otherwise, the replay code will ignore them. (BUG: ext4_fc_reserve_space() could violate this requirement if called with a len of blocksize - 9 or blocksize - 8. Fortunately, this doesn't seem to happen currently.) - tlvs must end at least 1 byte before the end of the block. Otherwise the replay code will consider them to be invalid. This quirk contributed to a bug (fixed by an earlier commit) where uninitialized memory was being leaked to disk in the last byte of blocks. Also, strangely these constraints don't apply to the replay code in e2fsprogs, which will accept any tlvs in the blocks (with no bounds checks at all, but that is a separate issue...). Given that this all seems to be a bug, let's fix it by just filling blocks with tlv entries in the natural way. Note that old kernels will be unable to replay fast-commit journals created by kernels that have this commit. Fixes: aa75f4d3daae ("ext4: main fast-commit commit path") Cc: <stable@vger.kernel.org> # v5.10+ Signed-off-by: Eric Biggers <ebiggers@google.com> Link: https://lore.kernel.org/r/20221106224841.279231-7-ebiggers@kernel.org Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-11-07 06:48:40 +08:00
end = start + journal->j_blocksize;
if (state->fc_replay_expected_off == 0) {
state->fc_cur_tag = 0;
state->fc_replay_num_tags = 0;
state->fc_crc = 0;
state->fc_regions = NULL;
state->fc_regions_valid = state->fc_regions_used =
state->fc_regions_size = 0;
/* Check if we can stop early */
if (le16_to_cpu(((struct ext4_fc_tl *)start)->fc_tag)
!= EXT4_FC_TAG_HEAD)
return 0;
}
if (off != state->fc_replay_expected_off) {
ret = -EFSCORRUPTED;
goto out_err;
}
state->fc_replay_expected_off++;
ext4: fix off-by-one errors in fast-commit block filling Due to several different off-by-one errors, or perhaps due to a late change in design that wasn't fully reflected in the code that was actually merged, there are several very strange constraints on how fast-commit blocks are filled with tlv entries: - tlvs must start at least 10 bytes before the end of the block, even though the minimum tlv length is 8. Otherwise, the replay code will ignore them. (BUG: ext4_fc_reserve_space() could violate this requirement if called with a len of blocksize - 9 or blocksize - 8. Fortunately, this doesn't seem to happen currently.) - tlvs must end at least 1 byte before the end of the block. Otherwise the replay code will consider them to be invalid. This quirk contributed to a bug (fixed by an earlier commit) where uninitialized memory was being leaked to disk in the last byte of blocks. Also, strangely these constraints don't apply to the replay code in e2fsprogs, which will accept any tlvs in the blocks (with no bounds checks at all, but that is a separate issue...). Given that this all seems to be a bug, let's fix it by just filling blocks with tlv entries in the natural way. Note that old kernels will be unable to replay fast-commit journals created by kernels that have this commit. Fixes: aa75f4d3daae ("ext4: main fast-commit commit path") Cc: <stable@vger.kernel.org> # v5.10+ Signed-off-by: Eric Biggers <ebiggers@google.com> Link: https://lore.kernel.org/r/20221106224841.279231-7-ebiggers@kernel.org Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-11-07 06:48:40 +08:00
for (cur = start; cur <= end - EXT4_FC_TAG_BASE_LEN;
cur = cur + EXT4_FC_TAG_BASE_LEN + tl.fc_len) {
ext4_fc_get_tl(&tl, cur);
val = cur + EXT4_FC_TAG_BASE_LEN;
if (tl.fc_len > end - val ||
!ext4_fc_value_len_isvalid(sbi, tl.fc_tag, tl.fc_len)) {
ret = state->fc_replay_num_tags ?
JBD2_FC_REPLAY_STOP : -ECANCELED;
goto out_err;
}
ext4_debug("Scan phase, tag:%s, blk %lld\n",
tag2str(tl.fc_tag), bh->b_blocknr);
switch (tl.fc_tag) {
case EXT4_FC_TAG_ADD_RANGE:
memcpy(&ext, val, sizeof(ext));
ex = (struct ext4_extent *)&ext.fc_ex;
ret = ext4_fc_record_regions(sb,
le32_to_cpu(ext.fc_ino),
le32_to_cpu(ex->ee_block), ext4_ext_pblock(ex),
ext4_ext_get_actual_len(ex), 0);
if (ret < 0)
break;
ret = JBD2_FC_REPLAY_CONTINUE;
fallthrough;
case EXT4_FC_TAG_DEL_RANGE:
case EXT4_FC_TAG_LINK:
case EXT4_FC_TAG_UNLINK:
case EXT4_FC_TAG_CREAT:
case EXT4_FC_TAG_INODE:
case EXT4_FC_TAG_PAD:
state->fc_cur_tag++;
state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,
EXT4_FC_TAG_BASE_LEN + tl.fc_len);
break;
case EXT4_FC_TAG_TAIL:
state->fc_cur_tag++;
memcpy(&tail, val, sizeof(tail));
state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,
EXT4_FC_TAG_BASE_LEN +
offsetof(struct ext4_fc_tail,
fc_crc));
if (le32_to_cpu(tail.fc_tid) == expected_tid &&
le32_to_cpu(tail.fc_crc) == state->fc_crc) {
state->fc_replay_num_tags = state->fc_cur_tag;
state->fc_regions_valid =
state->fc_regions_used;
} else {
ret = state->fc_replay_num_tags ?
JBD2_FC_REPLAY_STOP : -EFSBADCRC;
}
state->fc_crc = 0;
break;
case EXT4_FC_TAG_HEAD:
memcpy(&head, val, sizeof(head));
if (le32_to_cpu(head.fc_features) &
~EXT4_FC_SUPPORTED_FEATURES) {
ret = -EOPNOTSUPP;
break;
}
if (le32_to_cpu(head.fc_tid) != expected_tid) {
ret = JBD2_FC_REPLAY_STOP;
break;
}
state->fc_cur_tag++;
state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,
EXT4_FC_TAG_BASE_LEN + tl.fc_len);
break;
default:
ret = state->fc_replay_num_tags ?
JBD2_FC_REPLAY_STOP : -ECANCELED;
}
if (ret < 0 || ret == JBD2_FC_REPLAY_STOP)
break;
}
out_err:
trace_ext4_fc_replay_scan(sb, ret, off);
return ret;
}
/*
* Main recovery path entry point.
* The meaning of return codes is similar as above.
*/
static int ext4_fc_replay(journal_t *journal, struct buffer_head *bh,
enum passtype pass, int off, tid_t expected_tid)
{
struct super_block *sb = journal->j_private;
struct ext4_sb_info *sbi = EXT4_SB(sb);
struct ext4_fc_tl_mem tl;
__u8 *start, *end, *cur, *val;
int ret = JBD2_FC_REPLAY_CONTINUE;
struct ext4_fc_replay_state *state = &sbi->s_fc_replay_state;
struct ext4_fc_tail tail;
if (pass == PASS_SCAN) {
state->fc_current_pass = PASS_SCAN;
return ext4_fc_replay_scan(journal, bh, off, expected_tid);
}
if (state->fc_current_pass != pass) {
state->fc_current_pass = pass;
sbi->s_mount_state |= EXT4_FC_REPLAY;
}
if (!sbi->s_fc_replay_state.fc_replay_num_tags) {
ext4_debug("Replay stops\n");
ext4_fc_set_bitmaps_and_counters(sb);
return 0;
}
#ifdef CONFIG_EXT4_DEBUG
if (sbi->s_fc_debug_max_replay && off >= sbi->s_fc_debug_max_replay) {
pr_warn("Dropping fc block %d because max_replay set\n", off);
return JBD2_FC_REPLAY_STOP;
}
#endif
start = (u8 *)bh->b_data;
ext4: fix off-by-one errors in fast-commit block filling Due to several different off-by-one errors, or perhaps due to a late change in design that wasn't fully reflected in the code that was actually merged, there are several very strange constraints on how fast-commit blocks are filled with tlv entries: - tlvs must start at least 10 bytes before the end of the block, even though the minimum tlv length is 8. Otherwise, the replay code will ignore them. (BUG: ext4_fc_reserve_space() could violate this requirement if called with a len of blocksize - 9 or blocksize - 8. Fortunately, this doesn't seem to happen currently.) - tlvs must end at least 1 byte before the end of the block. Otherwise the replay code will consider them to be invalid. This quirk contributed to a bug (fixed by an earlier commit) where uninitialized memory was being leaked to disk in the last byte of blocks. Also, strangely these constraints don't apply to the replay code in e2fsprogs, which will accept any tlvs in the blocks (with no bounds checks at all, but that is a separate issue...). Given that this all seems to be a bug, let's fix it by just filling blocks with tlv entries in the natural way. Note that old kernels will be unable to replay fast-commit journals created by kernels that have this commit. Fixes: aa75f4d3daae ("ext4: main fast-commit commit path") Cc: <stable@vger.kernel.org> # v5.10+ Signed-off-by: Eric Biggers <ebiggers@google.com> Link: https://lore.kernel.org/r/20221106224841.279231-7-ebiggers@kernel.org Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-11-07 06:48:40 +08:00
end = start + journal->j_blocksize;
ext4: fix off-by-one errors in fast-commit block filling Due to several different off-by-one errors, or perhaps due to a late change in design that wasn't fully reflected in the code that was actually merged, there are several very strange constraints on how fast-commit blocks are filled with tlv entries: - tlvs must start at least 10 bytes before the end of the block, even though the minimum tlv length is 8. Otherwise, the replay code will ignore them. (BUG: ext4_fc_reserve_space() could violate this requirement if called with a len of blocksize - 9 or blocksize - 8. Fortunately, this doesn't seem to happen currently.) - tlvs must end at least 1 byte before the end of the block. Otherwise the replay code will consider them to be invalid. This quirk contributed to a bug (fixed by an earlier commit) where uninitialized memory was being leaked to disk in the last byte of blocks. Also, strangely these constraints don't apply to the replay code in e2fsprogs, which will accept any tlvs in the blocks (with no bounds checks at all, but that is a separate issue...). Given that this all seems to be a bug, let's fix it by just filling blocks with tlv entries in the natural way. Note that old kernels will be unable to replay fast-commit journals created by kernels that have this commit. Fixes: aa75f4d3daae ("ext4: main fast-commit commit path") Cc: <stable@vger.kernel.org> # v5.10+ Signed-off-by: Eric Biggers <ebiggers@google.com> Link: https://lore.kernel.org/r/20221106224841.279231-7-ebiggers@kernel.org Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2022-11-07 06:48:40 +08:00
for (cur = start; cur <= end - EXT4_FC_TAG_BASE_LEN;
cur = cur + EXT4_FC_TAG_BASE_LEN + tl.fc_len) {
ext4_fc_get_tl(&tl, cur);
val = cur + EXT4_FC_TAG_BASE_LEN;
if (state->fc_replay_num_tags == 0) {
ret = JBD2_FC_REPLAY_STOP;
ext4_fc_set_bitmaps_and_counters(sb);
break;
}
ext4_debug("Replay phase, tag:%s\n", tag2str(tl.fc_tag));
state->fc_replay_num_tags--;
switch (tl.fc_tag) {
case EXT4_FC_TAG_LINK:
ret = ext4_fc_replay_link(sb, &tl, val);
break;
case EXT4_FC_TAG_UNLINK:
ret = ext4_fc_replay_unlink(sb, &tl, val);
break;
case EXT4_FC_TAG_ADD_RANGE:
ret = ext4_fc_replay_add_range(sb, &tl, val);
break;
case EXT4_FC_TAG_CREAT:
ret = ext4_fc_replay_create(sb, &tl, val);
break;
case EXT4_FC_TAG_DEL_RANGE:
ret = ext4_fc_replay_del_range(sb, &tl, val);
break;
case EXT4_FC_TAG_INODE:
ret = ext4_fc_replay_inode(sb, &tl, val);
break;
case EXT4_FC_TAG_PAD:
trace_ext4_fc_replay(sb, EXT4_FC_TAG_PAD, 0,
tl.fc_len, 0);
break;
case EXT4_FC_TAG_TAIL:
trace_ext4_fc_replay(sb, EXT4_FC_TAG_TAIL,
0, tl.fc_len, 0);
memcpy(&tail, val, sizeof(tail));
WARN_ON(le32_to_cpu(tail.fc_tid) != expected_tid);
break;
case EXT4_FC_TAG_HEAD:
break;
default:
trace_ext4_fc_replay(sb, tl.fc_tag, 0, tl.fc_len, 0);
ret = -ECANCELED;
break;
}
if (ret < 0)
break;
ret = JBD2_FC_REPLAY_CONTINUE;
}
return ret;
}
void ext4_fc_init(struct super_block *sb, journal_t *journal)
{
/*
* We set replay callback even if fast commit disabled because we may
* could still have fast commit blocks that need to be replayed even if
* fast commit has now been turned off.
*/
journal->j_fc_replay_callback = ext4_fc_replay;
if (!test_opt2(sb, JOURNAL_FAST_COMMIT))
return;
journal->j_fc_cleanup_callback = ext4_fc_cleanup;
}
static const char * const fc_ineligible_reasons[] = {
[EXT4_FC_REASON_XATTR] = "Extended attributes changed",
[EXT4_FC_REASON_CROSS_RENAME] = "Cross rename",
[EXT4_FC_REASON_JOURNAL_FLAG_CHANGE] = "Journal flag changed",
[EXT4_FC_REASON_NOMEM] = "Insufficient memory",
[EXT4_FC_REASON_SWAP_BOOT] = "Swap boot",
[EXT4_FC_REASON_RESIZE] = "Resize",
[EXT4_FC_REASON_RENAME_DIR] = "Dir renamed",
[EXT4_FC_REASON_FALLOC_RANGE] = "Falloc range op",
[EXT4_FC_REASON_INODE_JOURNAL_DATA] = "Data journalling",
[EXT4_FC_REASON_ENCRYPTED_FILENAME] = "Encrypted filename",
};
int ext4_fc_info_show(struct seq_file *seq, void *v)
{
struct ext4_sb_info *sbi = EXT4_SB((struct super_block *)seq->private);
struct ext4_fc_stats *stats = &sbi->s_fc_stats;
int i;
if (v != SEQ_START_TOKEN)
return 0;
seq_printf(seq,
"fc stats:\n%ld commits\n%ld ineligible\n%ld numblks\n%lluus avg_commit_time\n",
stats->fc_num_commits, stats->fc_ineligible_commits,
stats->fc_numblks,
div_u64(stats->s_fc_avg_commit_time, 1000));
seq_puts(seq, "Ineligible reasons:\n");
for (i = 0; i < EXT4_FC_REASON_MAX; i++)
seq_printf(seq, "\"%s\":\t%d\n", fc_ineligible_reasons[i],
stats->fc_ineligible_reason_count[i]);
return 0;
}
int __init ext4_fc_init_dentry_cache(void)
{
ext4_fc_dentry_cachep = KMEM_CACHE(ext4_fc_dentry_update,
SLAB_RECLAIM_ACCOUNT);
if (ext4_fc_dentry_cachep == NULL)
return -ENOMEM;
return 0;
}
void ext4_fc_destroy_dentry_cache(void)
{
kmem_cache_destroy(ext4_fc_dentry_cachep);
}